github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-17 05:36:16 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #9048] 0.5.9 RC makes models spit out garbage #31649

New Issue
Closed
opened 2026-04-22 12:18:18 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-22 12:18:18 -05:00
Owner
Copy Link

Originally created by @TheSeraph on GitHub (Feb 12, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9048

What is the issue?

Trying 0.5.9 on windows, and models spit out utter garbage

`base) PS C:\WINDOWS\system32> ollama run llama3.3:70b-instruct-q2_K-64k

Test firing AVX support
(\\ " ( (. or2 GA ( ( ( in" ( (. " ( ( ( it as " ( ( ( (". to " ( " ( ( ( (2 ( ( ( " (. ( ( ( ( ( ( ( (". S2
( ( ( ( ( " ( ( " A ( ( O1 ( (2 ( ( ( ( " ( " a' ( ( ( B D ( ( ( ( ( "\\ ( (". a real ( ( ( it ( ( visit " ( (
( " ( LE A ( ( F ( ( ( ( ( ( ( ( ( ( " ( ( ( ( " ( ( ( ( " ( (2 " ( I ( (. ( ( ( " ( ( (2 ( ( ( ( (2 (".

Hello
\\\\ ( of\\' " \ ( ("\\ to ( (" it/. ". "\\\\ \ (\\\\.\\�\\) ". (\\ it\\
\.\\\\"\\ "\ "\\\. (\\\\\\.."

Hello are you alive?
"\\ ('\\\\\ or\\\\\\\\\\\\\\\ a2 (\\\\/\\ (.\ " (\\ (

exit
\\ " "\\\\\\\\\\ "\\\\ ( (\\\\\\\\\\/\\ ('\\"\\'\ \\\\\". (" \\\'
to\\\\\\ "\\\\\\

/exit
(base) PS C:\WINDOWS\system32> ollama run llama3.3:70b-instruct-q2_K
testing 1 2 3
\ " " a,\\, " " " " " \ " " n " " " " or " n" the " that n \ n & a, " or

/bye
(base) PS C:\WINDOWS\system32> ollama -v
ollama version is 0.5.9-rc0
(base) PS C:\WINDOWS\system32> `

Image

Relevant log output

2025/02/12 14:27:09 routes.go:1186: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:true OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE:Q4_0 OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:T:\\ai\\models\\ollama\\ OLLAMA_MULTIUSER_CACHE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[app://obsidian.md* http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]"
time=2025-02-12T14:27:09.310Z level=INFO source=images.go:432 msg="total blobs: 94"
time=2025-02-12T14:27:09.313Z level=INFO source=images.go:439 msg="total unused blobs removed: 0"
time=2025-02-12T14:27:09.314Z level=INFO source=routes.go:1237 msg="Listening on [::]:11434 (version 0.5.9-rc0)"
time=2025-02-12T14:27:09.314Z level=INFO source=gpu.go:217 msg="looking for compatible GPUs"
time=2025-02-12T14:27:09.314Z level=INFO source=gpu_windows.go:167 msg=packages count=1
time=2025-02-12T14:27:09.314Z level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16
time=2025-02-12T14:27:10.261Z level=INFO source=types.go:130 msg="inference compute" id=0 library=rocm variant="" compute=gfx1100 driver=6.2 name="AMD Radeon RX 7900 XTX" total="24.0 GiB" available="23.8 GiB"
time=2025-02-12T14:27:10.261Z level=INFO source=types.go:130 msg="inference compute" id=1 library=rocm variant="" compute=gfx1100 driver=6.2 name="AMD Radeon RX 7900 XTX" total="24.0 GiB" available="23.8 GiB"
[GIN] 2025/02/12 - 14:27:19 | 200 |            0s |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/12 - 14:27:19 | 200 |      5.8169ms |       127.0.0.1 | GET      "/api/tags"
[GIN] 2025/02/12 - 14:27:35 | 200 |            0s |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/12 - 14:27:35 | 200 |     54.6489ms |       127.0.0.1 | POST     "/api/show"
time=2025-02-12T14:27:36.324Z level=INFO source=sched.go:185 msg="one or more GPUs detected that are unable to accurately report free memory - disabling default concurrency"
time=2025-02-12T14:27:41.759Z level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="38.0 GiB" free_swap="28.3 GiB"
time=2025-02-12T14:27:42.541Z level=INFO source=memory.go:356 msg="offload to rocm" layers.requested=-1 layers.model=81 layers.offload=80 layers.split=40,40 memory.available="[23.7 GiB 23.7 GiB]" memory.gpu_overhead="0 B" memory.required.full="47.7 GiB" memory.required.partial="46.9 GiB" memory.required.kv="5.0 GiB" memory.required.allocations="[23.5 GiB 23.5 GiB]" memory.weights.total="28.4 GiB" memory.weights.repeating="27.6 GiB" memory.weights.nonrepeating="822.0 MiB" memory.graph.full="8.4 GiB" memory.graph.partial="8.4 GiB"
time=2025-02-12T14:27:42.541Z level=INFO source=server.go:185 msg="enabling flash attention"
time=2025-02-12T14:27:42.548Z level=INFO source=server.go:381 msg="starting llama server" cmd="T:\\ai\\ollama\\ollama.exe runner --model T:\\ai\\models\\ollama\\blobs\\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 --ctx-size 65535 --batch-size 512 --n-gpu-layers 80 --threads 8 --flash-attn --kv-cache-type q4_0 --parallel 1 --tensor-split 40,40 --port 39854"
time=2025-02-12T14:27:42.553Z level=INFO source=sched.go:449 msg="loaded runners" count=1
time=2025-02-12T14:27:42.553Z level=INFO source=server.go:558 msg="waiting for llama runner to start responding"
time=2025-02-12T14:27:42.554Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server error"
time=2025-02-12T14:27:42.583Z level=INFO source=runner.go:936 msg="starting go runner"
time=2025-02-12T14:27:42.584Z level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=8
time=2025-02-12T14:27:42.585Z level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:39854"
time=2025-02-12T14:27:42.804Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server loading model"
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 2 ROCm devices:
  Device 0: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no
  Device 1: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no
load_backend: loaded ROCm backend from T:\ai\ollama\lib\ollama\rocm\ggml-hip.dll
load_backend: loaded CPU backend from T:\ai\ollama\lib\ollama\ggml-cpu-haswell.dll
llama_load_model_from_file: using device ROCm0 (AMD Radeon RX 7900 XTX) - 24194 MiB free
llama_load_model_from_file: using device ROCm1 (AMD Radeon RX 7900 XTX) - 24411 MiB free
llama_model_loader: loaded meta data with 36 key-value pairs and 724 tensors from T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = llama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Llama 3.1 70B Instruct 2024 12
llama_model_loader: - kv   3:                            general.version str              = 2024-12
llama_model_loader: - kv   4:                           general.finetune str              = Instruct
llama_model_loader: - kv   5:                           general.basename str              = Llama-3.1
llama_model_loader: - kv   6:                         general.size_label str              = 70B
llama_model_loader: - kv   7:                            general.license str              = llama3.1
llama_model_loader: - kv   8:                   general.base_model.count u32              = 1
llama_model_loader: - kv   9:                  general.base_model.0.name str              = Llama 3.1 70B
llama_model_loader: - kv  10:          general.base_model.0.organization str              = Meta Llama
llama_model_loader: - kv  11:              general.base_model.0.repo_url str              = https://huggingface.co/meta-llama/Lla...
llama_model_loader: - kv  12:                               general.tags arr[str,5]       = ["facebook", "meta", "pytorch", "llam...
llama_model_loader: - kv  13:                          general.languages arr[str,7]       = ["fr", "it", "pt", "hi", "es", "th", ...
llama_model_loader: - kv  14:                          llama.block_count u32              = 80
llama_model_loader: - kv  15:                       llama.context_length u32              = 131072
llama_model_loader: - kv  16:                     llama.embedding_length u32              = 8192
llama_model_loader: - kv  17:                  llama.feed_forward_length u32              = 28672
llama_model_loader: - kv  18:                 llama.attention.head_count u32              = 64
llama_model_loader: - kv  19:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  20:                       llama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  21:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  22:                 llama.attention.key_length u32              = 128
llama_model_loader: - kv  23:               llama.attention.value_length u32              = 128
llama_model_loader: - kv  24:                          general.file_type u32              = 10
llama_model_loader: - kv  25:                           llama.vocab_size u32              = 128256
llama_model_loader: - kv  26:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv  27:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  28:                         tokenizer.ggml.pre str              = llama-bpe
llama_model_loader: - kv  29:                      tokenizer.ggml.tokens arr[str,128256]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  30:                  tokenizer.ggml.token_type arr[i32,128256]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  31:                      tokenizer.ggml.merges arr[str,280147]  = ["Ä  Ä ", "Ä  Ä Ä Ä ", "Ä Ä  Ä Ä ", "...
llama_model_loader: - kv  32:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  33:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {{- bos_token }}\n{%- if custom_tools ...
llama_model_loader: - kv  35:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  162 tensors
llama_model_loader: - type q2_K:  321 tensors
llama_model_loader: - type q3_K:  160 tensors
llama_model_loader: - type q5_K:   80 tensors
llama_model_loader: - type q6_K:    1 tensors
llm_load_vocab: special tokens cache size = 256
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 8192
llm_load_print_meta: n_layer          = 80
llm_load_print_meta: n_head           = 64
llm_load_print_meta: n_head_kv        = 8
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 8
llm_load_print_meta: n_embd_k_gqa     = 1024
llm_load_print_meta: n_embd_v_gqa     = 1024
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 28672
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 0
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 500000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 131072
llm_load_print_meta: rope_finetuned   = unknown
llm_load_print_meta: ssm_d_conv       = 0
llm_load_print_meta: ssm_d_inner      = 0
llm_load_print_meta: ssm_d_state      = 0
llm_load_print_meta: ssm_dt_rank      = 0
llm_load_print_meta: ssm_dt_b_c_rms   = 0
llm_load_print_meta: model type       = 70B
llm_load_print_meta: model ftype      = Q2_K - Medium
llm_load_print_meta: model params     = 70.55 B
llm_load_print_meta: model size       = 24.56 GiB (2.99 BPW)
llm_load_print_meta: general.name     = Llama 3.1 70B Instruct 2024 12
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOM token        = 128008 '<|eom_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOG token        = 128008 '<|eom_id|>'
llm_load_print_meta: EOG token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
llm_load_tensors: offloading 80 repeating layers to GPU
llm_load_tensors: offloaded 80/81 layers to GPU
llm_load_tensors:   CPU_Mapped model buffer size = 25145.77 MiB
llm_load_tensors:        ROCm0 model buffer size = 11997.50 MiB
llm_load_tensors:        ROCm1 model buffer size = 11997.50 MiB
llama_new_context_with_model: n_seq_max     = 1
llama_new_context_with_model: n_ctx         = 65536
llama_new_context_with_model: n_ctx_per_seq = 65536
llama_new_context_with_model: n_batch       = 512
llama_new_context_with_model: n_ubatch      = 512
llama_new_context_with_model: flash_attn    = 1
llama_new_context_with_model: freq_base     = 500000.0
llama_new_context_with_model: freq_scale    = 1
llama_new_context_with_model: n_ctx_per_seq (65536) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 65536, offload = 1, type_k = 'q4_0', type_v = 'q4_0', n_layer = 80, can_shift = 1
llama_kv_cache_init:      ROCm0 KV buffer size =  2880.00 MiB
llama_kv_cache_init:      ROCm1 KV buffer size =  2880.00 MiB
llama_new_context_with_model: KV self size  = 5760.00 MiB, K (q4_0): 2880.00 MiB, V (q4_0): 2880.00 MiB
llama_new_context_with_model:        CPU  output buffer size =     0.52 MiB
llama_new_context_with_model:      ROCm0 compute buffer size =  1088.45 MiB
llama_new_context_with_model:      ROCm1 compute buffer size =   208.00 MiB
llama_new_context_with_model:  ROCm_Host compute buffer size =   144.01 MiB
llama_new_context_with_model: graph nodes  = 2247
llama_new_context_with_model: graph splits = 5 (with bs=512), 4 (with bs=1)
time=2025-02-12T14:28:24.861Z level=INFO source=server.go:597 msg="llama runner started in 42.31 seconds"
[GIN] 2025/02/12 - 14:28:24 | 200 |   49.3998048s |       127.0.0.1 | POST     "/api/generate"
[GIN] 2025/02/12 - 14:29:01 | 200 |   16.1885007s |       127.0.0.1 | POST     "/api/chat"
llama_model_loader: loaded meta data with 36 key-value pairs and 724 tensors from T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = llama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Llama 3.1 70B Instruct 2024 12
llama_model_loader: - kv   3:                            general.version str              = 2024-12
llama_model_loader: - kv   4:                           general.finetune str              = Instruct
llama_model_loader: - kv   5:                           general.basename str              = Llama-3.1
llama_model_loader: - kv   6:                         general.size_label str              = 70B
llama_model_loader: - kv   7:                            general.license str              = llama3.1
llama_model_loader: - kv   8:                   general.base_model.count u32              = 1
llama_model_loader: - kv   9:                  general.base_model.0.name str              = Llama 3.1 70B
llama_model_loader: - kv  10:          general.base_model.0.organization str              = Meta Llama
llama_model_loader: - kv  11:              general.base_model.0.repo_url str              = https://huggingface.co/meta-llama/Lla...
llama_model_loader: - kv  12:                               general.tags arr[str,5]       = ["facebook", "meta", "pytorch", "llam...
llama_model_loader: - kv  13:                          general.languages arr[str,7]       = ["fr", "it", "pt", "hi", "es", "th", ...
llama_model_loader: - kv  14:                          llama.block_count u32              = 80
llama_model_loader: - kv  15:                       llama.context_length u32              = 131072
llama_model_loader: - kv  16:                     llama.embedding_length u32              = 8192
llama_model_loader: - kv  17:                  llama.feed_forward_length u32              = 28672
llama_model_loader: - kv  18:                 llama.attention.head_count u32              = 64
llama_model_loader: - kv  19:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  20:                       llama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  21:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  22:                 llama.attention.key_length u32              = 128
llama_model_loader: - kv  23:               llama.attention.value_length u32              = 128
llama_model_loader: - kv  24:                          general.file_type u32              = 10
llama_model_loader: - kv  25:                           llama.vocab_size u32              = 128256
llama_model_loader: - kv  26:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv  27:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  28:                         tokenizer.ggml.pre str              = llama-bpe
llama_model_loader: - kv  29:                      tokenizer.ggml.tokens arr[str,128256]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  30:                  tokenizer.ggml.token_type arr[i32,128256]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  31:                      tokenizer.ggml.merges arr[str,280147]  = ["Ä  Ä ", "Ä  Ä Ä Ä ", "Ä Ä  Ä Ä ", "...
llama_model_loader: - kv  32:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  33:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {{- bos_token }}\n{%- if custom_tools ...
llama_model_loader: - kv  35:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  162 tensors
llama_model_loader: - type q2_K:  321 tensors
llama_model_loader: - type q3_K:  160 tensors
llama_model_loader: - type q5_K:   80 tensors
llama_model_loader: - type q6_K:    1 tensors
llm_load_vocab: special tokens cache size = 256
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 1
llm_load_print_meta: model type       = ?B
llm_load_print_meta: model ftype      = all F32
llm_load_print_meta: model params     = 70.55 B
llm_load_print_meta: model size       = 24.56 GiB (2.99 BPW)
llm_load_print_meta: general.name     = Llama 3.1 70B Instruct 2024 12
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOM token        = 128008 '<|eom_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOG token        = 128008 '<|eom_id|>'
llm_load_print_meta: EOG token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
llama_model_load: vocab only - skipping tensors
[GIN] 2025/02/12 - 14:29:12 | 200 |    9.5654491s |       127.0.0.1 | POST     "/api/chat"
[GIN] 2025/02/12 - 14:29:21 | 200 |    3.4621003s |       127.0.0.1 | POST     "/api/chat"
[GIN] 2025/02/12 - 14:29:28 | 200 |    4.4678599s |       127.0.0.1 | POST     "/api/chat"
[GIN] 2025/02/12 - 14:29:47 | 200 |            0s |       127.0.0.1 | HEAD     "/"
[GIN] 2025/02/12 - 14:29:47 | 200 |     23.6152ms |       127.0.0.1 | POST     "/api/show"
time=2025-02-12T14:29:54.457Z level=INFO source=sched.go:730 msg="new model will fit in available VRAM, loading" model=T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 library=rocm parallel=4 required="28.6 GiB"
time=2025-02-12T14:29:55.232Z level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="38.2 GiB" free_swap="26.1 GiB"
time=2025-02-12T14:29:56.034Z level=INFO source=memory.go:356 msg="offload to rocm" layers.requested=-1 layers.model=81 layers.offload=81 layers.split=41,40 memory.available="[22.4 GiB 22.4 GiB]" memory.gpu_overhead="0 B" memory.required.full="28.6 GiB" memory.required.partial="28.6 GiB" memory.required.kv="640.0 MiB" memory.required.allocations="[14.7 GiB 13.9 GiB]" memory.weights.total="24.1 GiB" memory.weights.repeating="23.3 GiB" memory.weights.nonrepeating="822.0 MiB" memory.graph.full="1.1 GiB" memory.graph.partial="1.1 GiB"
time=2025-02-12T14:29:56.035Z level=INFO source=server.go:185 msg="enabling flash attention"
time=2025-02-12T14:29:56.039Z level=INFO source=server.go:381 msg="starting llama server" cmd="T:\\ai\\ollama\\ollama.exe runner --model T:\\ai\\models\\ollama\\blobs\\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 --ctx-size 8192 --batch-size 512 --n-gpu-layers 81 --threads 8 --flash-attn --kv-cache-type q4_0 --parallel 4 --tensor-split 41,40 --port 40130"
time=2025-02-12T14:29:56.044Z level=INFO source=sched.go:449 msg="loaded runners" count=1
time=2025-02-12T14:29:56.044Z level=INFO source=server.go:558 msg="waiting for llama runner to start responding"
time=2025-02-12T14:29:56.045Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server error"
time=2025-02-12T14:29:56.080Z level=INFO source=runner.go:936 msg="starting go runner"
time=2025-02-12T14:29:56.081Z level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=8
time=2025-02-12T14:29:56.082Z level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:40130"
time=2025-02-12T14:29:56.296Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server loading model"
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 2 ROCm devices:
  Device 0: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no
  Device 1: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no
load_backend: loaded ROCm backend from T:\ai\ollama\lib\ollama\rocm\ggml-hip.dll
load_backend: loaded CPU backend from T:\ai\ollama\lib\ollama\ggml-cpu-haswell.dll
llama_load_model_from_file: using device ROCm0 (AMD Radeon RX 7900 XTX) - 24194 MiB free
llama_load_model_from_file: using device ROCm1 (AMD Radeon RX 7900 XTX) - 24411 MiB free
llama_model_loader: loaded meta data with 36 key-value pairs and 724 tensors from T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = llama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Llama 3.1 70B Instruct 2024 12
llama_model_loader: - kv   3:                            general.version str              = 2024-12
llama_model_loader: - kv   4:                           general.finetune str              = Instruct
llama_model_loader: - kv   5:                           general.basename str              = Llama-3.1
llama_model_loader: - kv   6:                         general.size_label str              = 70B
llama_model_loader: - kv   7:                            general.license str              = llama3.1
llama_model_loader: - kv   8:                   general.base_model.count u32              = 1
llama_model_loader: - kv   9:                  general.base_model.0.name str              = Llama 3.1 70B
llama_model_loader: - kv  10:          general.base_model.0.organization str              = Meta Llama
llama_model_loader: - kv  11:              general.base_model.0.repo_url str              = https://huggingface.co/meta-llama/Lla...
llama_model_loader: - kv  12:                               general.tags arr[str,5]       = ["facebook", "meta", "pytorch", "llam...
llama_model_loader: - kv  13:                          general.languages arr[str,7]       = ["fr", "it", "pt", "hi", "es", "th", ...
llama_model_loader: - kv  14:                          llama.block_count u32              = 80
llama_model_loader: - kv  15:                       llama.context_length u32              = 131072
llama_model_loader: - kv  16:                     llama.embedding_length u32              = 8192
llama_model_loader: - kv  17:                  llama.feed_forward_length u32              = 28672
llama_model_loader: - kv  18:                 llama.attention.head_count u32              = 64
llama_model_loader: - kv  19:              llama.attention.head_count_kv u32              = 8
llama_model_loader: - kv  20:                       llama.rope.freq_base f32              = 500000.000000
llama_model_loader: - kv  21:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  22:                 llama.attention.key_length u32              = 128
llama_model_loader: - kv  23:               llama.attention.value_length u32              = 128
llama_model_loader: - kv  24:                          general.file_type u32              = 10
llama_model_loader: - kv  25:                           llama.vocab_size u32              = 128256
llama_model_loader: - kv  26:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv  27:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  28:                         tokenizer.ggml.pre str              = llama-bpe
llama_model_loader: - kv  29:                      tokenizer.ggml.tokens arr[str,128256]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  30:                  tokenizer.ggml.token_type arr[i32,128256]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  31:                      tokenizer.ggml.merges arr[str,280147]  = ["Ä  Ä ", "Ä  Ä Ä Ä ", "Ä Ä  Ä Ä ", "...
llama_model_loader: - kv  32:                tokenizer.ggml.bos_token_id u32              = 128000
llama_model_loader: - kv  33:                tokenizer.ggml.eos_token_id u32              = 128009
llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {{- bos_token }}\n{%- if custom_tools ...
llama_model_loader: - kv  35:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  162 tensors
llama_model_loader: - type q2_K:  321 tensors
llama_model_loader: - type q3_K:  160 tensors
llama_model_loader: - type q5_K:   80 tensors
llama_model_loader: - type q6_K:    1 tensors
llm_load_vocab: special tokens cache size = 256
llm_load_vocab: token to piece cache size = 0.7999 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = BPE
llm_load_print_meta: n_vocab          = 128256
llm_load_print_meta: n_merges         = 280147
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 8192
llm_load_print_meta: n_layer          = 80
llm_load_print_meta: n_head           = 64
llm_load_print_meta: n_head_kv        = 8
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 8
llm_load_print_meta: n_embd_k_gqa     = 1024
llm_load_print_meta: n_embd_v_gqa     = 1024
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 28672
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 0
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 500000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 131072
llm_load_print_meta: rope_finetuned   = unknown
llm_load_print_meta: ssm_d_conv       = 0
llm_load_print_meta: ssm_d_inner      = 0
llm_load_print_meta: ssm_d_state      = 0
llm_load_print_meta: ssm_dt_rank      = 0
llm_load_print_meta: ssm_dt_b_c_rms   = 0
llm_load_print_meta: model type       = 70B
llm_load_print_meta: model ftype      = Q2_K - Medium
llm_load_print_meta: model params     = 70.55 B
llm_load_print_meta: model size       = 24.56 GiB (2.99 BPW)
llm_load_print_meta: general.name     = Llama 3.1 70B Instruct 2024 12
llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
llm_load_print_meta: EOM token        = 128008 '<|eom_id|>'
llm_load_print_meta: LF token         = 128 'Ä'
llm_load_print_meta: EOG token        = 128008 '<|eom_id|>'
llm_load_print_meta: EOG token        = 128009 '<|eot_id|>'
llm_load_print_meta: max token length = 256
llm_load_tensors: offloading 80 repeating layers to GPU
llm_load_tensors: offloading output layer to GPU
llm_load_tensors: offloaded 81/81 layers to GPU
llm_load_tensors:        ROCm0 model buffer size = 12297.44 MiB
llm_load_tensors:        ROCm1 model buffer size = 12519.55 MiB
llm_load_tensors:   CPU_Mapped model buffer size =   328.78 MiB
llama_new_context_with_model: n_seq_max     = 4
llama_new_context_with_model: n_ctx         = 8192
llama_new_context_with_model: n_ctx_per_seq = 2048
llama_new_context_with_model: n_batch       = 2048
llama_new_context_with_model: n_ubatch      = 512
llama_new_context_with_model: flash_attn    = 1
llama_new_context_with_model: freq_base     = 500000.0
llama_new_context_with_model: freq_scale    = 1
llama_new_context_with_model: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 8192, offload = 1, type_k = 'q4_0', type_v = 'q4_0', n_layer = 80, can_shift = 1
llama_kv_cache_init:      ROCm0 KV buffer size =   369.00 MiB
llama_kv_cache_init:      ROCm1 KV buffer size =   351.00 MiB
llama_new_context_with_model: KV self size  =  720.00 MiB, K (q4_0):  360.00 MiB, V (q4_0):  360.00 MiB
llama_new_context_with_model:  ROCm_Host  output buffer size =     2.08 MiB
llama_new_context_with_model: pipeline parallelism enabled (n_copies=4)
llama_new_context_with_model:      ROCm0 compute buffer size =   296.01 MiB
llama_new_context_with_model:      ROCm1 compute buffer size =   362.52 MiB
llama_new_context_with_model:  ROCm_Host compute buffer size =    80.02 MiB
llama_new_context_with_model: graph nodes  = 2247
llama_new_context_with_model: graph splits = 3
time=2025-02-12T14:30:35.095Z level=INFO source=server.go:597 msg="llama runner started in 39.05 seconds"
[GIN] 2025/02/12 - 14:30:35 | 200 |   47.7650191s |       127.0.0.1 | POST     "/api/generate"
[GIN] 2025/02/12 - 14:30:44 | 200 |      2.87911s |       127.0.0.1 | POST     "/api/chat"
[GIN] 2025/02/12 - 14:31:30 | 200 |            0s |       127.0.0.1 | GET      "/api/version"
(base) PS C:\WINDOWS\system32>

OS

Windows

GPU

AMD

CPU

AMD

Ollama version

0.5.9

Originally created by @TheSeraph on GitHub (Feb 12, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9048 ### What is the issue? Trying 0.5.9 on windows, and models spit out utter garbage `base) PS C:\WINDOWS\system32> ollama run llama3.3:70b-instruct-q2_K-64k >>> Test firing AVX support (\\\\ " ( (. or2 GA ( ( ( in" ( (. " ( ( ( it as " ( ( ( (". to " ( \" ( ( ( (2 ( ( ( \" (. ( ( ( ( ( ( ( (". S2 ( ( ( ( ( " ( ( \" A ( ( O1 ( (2 ( ( ( ( " ( " a\' ( ( ( B D ( ( ( ( ( "\\\\ ( (". a real ( ( ( it ( ( visit " ( ( ( " ( LE A ( ( F ( ( ( ( ( ( ( ( ( ( " ( ( ( ( " ( ( ( ( " ( (2 " ( I ( (. ( ( ( " ( ( (2 ( ( ( ( (2 (". >>> Hello \\\\\\\\ ( of\\\\\' \" \ ( (\"\\\\ to ( (\" it\/. "\. \"\\\\\\\\ \\ (\\\\\\\\.\\\\�\\\\) \". (\\\\ it\\\\ \\.\\\\\\\\"\\\\ \"\\ "\\\\\\. (\\\\\\\\\\\\.." >>> Hello are you alive? "\\\\ (\'\\\\\\\\\ or\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ a2 (\\\\\\\\\/\\\\ (.\ " (\\\\ ( >>> exit \\\\ \" "\\\\\\\\\\\\\\\\\\\\ \"\\\\\\\\ ( (\\\\\\\\\\\\\\\\\\\\\/\\\\ (\'\\\\\"\\\\\'\\ \\\\\\\\\\". (\" \\\\\\' to\\\\\\\\\\\\ \"\\\\\\\\\\\\ >>> >>> /exit (base) PS C:\WINDOWS\system32> ollama run llama3.3:70b-instruct-q2_K >>> testing 1 2 3 \ " " a,\\\\, " " " " " \ " " n " " " " or " n\" the " that n \ n & a, " or >>> /bye (base) PS C:\WINDOWS\system32> ollama -v ollama version is 0.5.9-rc0 (base) PS C:\WINDOWS\system32> ` ![Image](https://github.com/user-attachments/assets/a3a18c01-712b-44a4-9009-5a0e8f474277) ### Relevant log output ```shell 2025/02/12 14:27:09 routes.go:1186: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:true OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://0.0.0.0:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE:Q4_0 OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:T:\\ai\\models\\ollama\\ OLLAMA_MULTIUSER_CACHE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[app://obsidian.md* http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]" time=2025-02-12T14:27:09.310Z level=INFO source=images.go:432 msg="total blobs: 94" time=2025-02-12T14:27:09.313Z level=INFO source=images.go:439 msg="total unused blobs removed: 0" time=2025-02-12T14:27:09.314Z level=INFO source=routes.go:1237 msg="Listening on [::]:11434 (version 0.5.9-rc0)" time=2025-02-12T14:27:09.314Z level=INFO source=gpu.go:217 msg="looking for compatible GPUs" time=2025-02-12T14:27:09.314Z level=INFO source=gpu_windows.go:167 msg=packages count=1 time=2025-02-12T14:27:09.314Z level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16 time=2025-02-12T14:27:10.261Z level=INFO source=types.go:130 msg="inference compute" id=0 library=rocm variant="" compute=gfx1100 driver=6.2 name="AMD Radeon RX 7900 XTX" total="24.0 GiB" available="23.8 GiB" time=2025-02-12T14:27:10.261Z level=INFO source=types.go:130 msg="inference compute" id=1 library=rocm variant="" compute=gfx1100 driver=6.2 name="AMD Radeon RX 7900 XTX" total="24.0 GiB" available="23.8 GiB" [GIN] 2025/02/12 - 14:27:19 | 200 | 0s | 127.0.0.1 | HEAD "/" [GIN] 2025/02/12 - 14:27:19 | 200 | 5.8169ms | 127.0.0.1 | GET "/api/tags" [GIN] 2025/02/12 - 14:27:35 | 200 | 0s | 127.0.0.1 | HEAD "/" [GIN] 2025/02/12 - 14:27:35 | 200 | 54.6489ms | 127.0.0.1 | POST "/api/show" time=2025-02-12T14:27:36.324Z level=INFO source=sched.go:185 msg="one or more GPUs detected that are unable to accurately report free memory - disabling default concurrency" time=2025-02-12T14:27:41.759Z level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="38.0 GiB" free_swap="28.3 GiB" time=2025-02-12T14:27:42.541Z level=INFO source=memory.go:356 msg="offload to rocm" layers.requested=-1 layers.model=81 layers.offload=80 layers.split=40,40 memory.available="[23.7 GiB 23.7 GiB]" memory.gpu_overhead="0 B" memory.required.full="47.7 GiB" memory.required.partial="46.9 GiB" memory.required.kv="5.0 GiB" memory.required.allocations="[23.5 GiB 23.5 GiB]" memory.weights.total="28.4 GiB" memory.weights.repeating="27.6 GiB" memory.weights.nonrepeating="822.0 MiB" memory.graph.full="8.4 GiB" memory.graph.partial="8.4 GiB" time=2025-02-12T14:27:42.541Z level=INFO source=server.go:185 msg="enabling flash attention" time=2025-02-12T14:27:42.548Z level=INFO source=server.go:381 msg="starting llama server" cmd="T:\\ai\\ollama\\ollama.exe runner --model T:\\ai\\models\\ollama\\blobs\\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 --ctx-size 65535 --batch-size 512 --n-gpu-layers 80 --threads 8 --flash-attn --kv-cache-type q4_0 --parallel 1 --tensor-split 40,40 --port 39854" time=2025-02-12T14:27:42.553Z level=INFO source=sched.go:449 msg="loaded runners" count=1 time=2025-02-12T14:27:42.553Z level=INFO source=server.go:558 msg="waiting for llama runner to start responding" time=2025-02-12T14:27:42.554Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server error" time=2025-02-12T14:27:42.583Z level=INFO source=runner.go:936 msg="starting go runner" time=2025-02-12T14:27:42.584Z level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=8 time=2025-02-12T14:27:42.585Z level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:39854" time=2025-02-12T14:27:42.804Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server loading model" ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no ggml_cuda_init: found 2 ROCm devices: Device 0: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no Device 1: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no load_backend: loaded ROCm backend from T:\ai\ollama\lib\ollama\rocm\ggml-hip.dll load_backend: loaded CPU backend from T:\ai\ollama\lib\ollama\ggml-cpu-haswell.dll llama_load_model_from_file: using device ROCm0 (AMD Radeon RX 7900 XTX) - 24194 MiB free llama_load_model_from_file: using device ROCm1 (AMD Radeon RX 7900 XTX) - 24411 MiB free llama_model_loader: loaded meta data with 36 key-value pairs and 724 tensors from T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 (version GGUF V3 (latest)) llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. llama_model_loader: - kv 0: general.architecture str = llama llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Llama 3.1 70B Instruct 2024 12 llama_model_loader: - kv 3: general.version str = 2024-12 llama_model_loader: - kv 4: general.finetune str = Instruct llama_model_loader: - kv 5: general.basename str = Llama-3.1 llama_model_loader: - kv 6: general.size_label str = 70B llama_model_loader: - kv 7: general.license str = llama3.1 llama_model_loader: - kv 8: general.base_model.count u32 = 1 llama_model_loader: - kv 9: general.base_model.0.name str = Llama 3.1 70B llama_model_loader: - kv 10: general.base_model.0.organization str = Meta Llama llama_model_loader: - kv 11: general.base_model.0.repo_url str = https://huggingface.co/meta-llama/Lla... llama_model_loader: - kv 12: general.tags arr[str,5] = ["facebook", "meta", "pytorch", "llam... llama_model_loader: - kv 13: general.languages arr[str,7] = ["fr", "it", "pt", "hi", "es", "th", ... llama_model_loader: - kv 14: llama.block_count u32 = 80 llama_model_loader: - kv 15: llama.context_length u32 = 131072 llama_model_loader: - kv 16: llama.embedding_length u32 = 8192 llama_model_loader: - kv 17: llama.feed_forward_length u32 = 28672 llama_model_loader: - kv 18: llama.attention.head_count u32 = 64 llama_model_loader: - kv 19: llama.attention.head_count_kv u32 = 8 llama_model_loader: - kv 20: llama.rope.freq_base f32 = 500000.000000 llama_model_loader: - kv 21: llama.attention.layer_norm_rms_epsilon f32 = 0.000010 llama_model_loader: - kv 22: llama.attention.key_length u32 = 128 llama_model_loader: - kv 23: llama.attention.value_length u32 = 128 llama_model_loader: - kv 24: general.file_type u32 = 10 llama_model_loader: - kv 25: llama.vocab_size u32 = 128256 llama_model_loader: - kv 26: llama.rope.dimension_count u32 = 128 llama_model_loader: - kv 27: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 28: tokenizer.ggml.pre str = llama-bpe llama_model_loader: - kv 29: tokenizer.ggml.tokens arr[str,128256] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 30: tokenizer.ggml.token_type arr[i32,128256] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 31: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "... llama_model_loader: - kv 32: tokenizer.ggml.bos_token_id u32 = 128000 llama_model_loader: - kv 33: tokenizer.ggml.eos_token_id u32 = 128009 llama_model_loader: - kv 34: tokenizer.chat_template str = {{- bos_token }}\n{%- if custom_tools ... llama_model_loader: - kv 35: general.quantization_version u32 = 2 llama_model_loader: - type f32: 162 tensors llama_model_loader: - type q2_K: 321 tensors llama_model_loader: - type q3_K: 160 tensors llama_model_loader: - type q5_K: 80 tensors llama_model_loader: - type q6_K: 1 tensors llm_load_vocab: special tokens cache size = 256 llm_load_vocab: token to piece cache size = 0.7999 MB llm_load_print_meta: format = GGUF V3 (latest) llm_load_print_meta: arch = llama llm_load_print_meta: vocab type = BPE llm_load_print_meta: n_vocab = 128256 llm_load_print_meta: n_merges = 280147 llm_load_print_meta: vocab_only = 0 llm_load_print_meta: n_ctx_train = 131072 llm_load_print_meta: n_embd = 8192 llm_load_print_meta: n_layer = 80 llm_load_print_meta: n_head = 64 llm_load_print_meta: n_head_kv = 8 llm_load_print_meta: n_rot = 128 llm_load_print_meta: n_swa = 0 llm_load_print_meta: n_embd_head_k = 128 llm_load_print_meta: n_embd_head_v = 128 llm_load_print_meta: n_gqa = 8 llm_load_print_meta: n_embd_k_gqa = 1024 llm_load_print_meta: n_embd_v_gqa = 1024 llm_load_print_meta: f_norm_eps = 0.0e+00 llm_load_print_meta: f_norm_rms_eps = 1.0e-05 llm_load_print_meta: f_clamp_kqv = 0.0e+00 llm_load_print_meta: f_max_alibi_bias = 0.0e+00 llm_load_print_meta: f_logit_scale = 0.0e+00 llm_load_print_meta: n_ff = 28672 llm_load_print_meta: n_expert = 0 llm_load_print_meta: n_expert_used = 0 llm_load_print_meta: causal attn = 1 llm_load_print_meta: pooling type = 0 llm_load_print_meta: rope type = 0 llm_load_print_meta: rope scaling = linear llm_load_print_meta: freq_base_train = 500000.0 llm_load_print_meta: freq_scale_train = 1 llm_load_print_meta: n_ctx_orig_yarn = 131072 llm_load_print_meta: rope_finetuned = unknown llm_load_print_meta: ssm_d_conv = 0 llm_load_print_meta: ssm_d_inner = 0 llm_load_print_meta: ssm_d_state = 0 llm_load_print_meta: ssm_dt_rank = 0 llm_load_print_meta: ssm_dt_b_c_rms = 0 llm_load_print_meta: model type = 70B llm_load_print_meta: model ftype = Q2_K - Medium llm_load_print_meta: model params = 70.55 B llm_load_print_meta: model size = 24.56 GiB (2.99 BPW) llm_load_print_meta: general.name = Llama 3.1 70B Instruct 2024 12 llm_load_print_meta: BOS token = 128000 '<|begin_of_text|>' llm_load_print_meta: EOS token = 128009 '<|eot_id|>' llm_load_print_meta: EOT token = 128009 '<|eot_id|>' llm_load_print_meta: EOM token = 128008 '<|eom_id|>' llm_load_print_meta: LF token = 128 'Ä' llm_load_print_meta: EOG token = 128008 '<|eom_id|>' llm_load_print_meta: EOG token = 128009 '<|eot_id|>' llm_load_print_meta: max token length = 256 llm_load_tensors: offloading 80 repeating layers to GPU llm_load_tensors: offloaded 80/81 layers to GPU llm_load_tensors: CPU_Mapped model buffer size = 25145.77 MiB llm_load_tensors: ROCm0 model buffer size = 11997.50 MiB llm_load_tensors: ROCm1 model buffer size = 11997.50 MiB llama_new_context_with_model: n_seq_max = 1 llama_new_context_with_model: n_ctx = 65536 llama_new_context_with_model: n_ctx_per_seq = 65536 llama_new_context_with_model: n_batch = 512 llama_new_context_with_model: n_ubatch = 512 llama_new_context_with_model: flash_attn = 1 llama_new_context_with_model: freq_base = 500000.0 llama_new_context_with_model: freq_scale = 1 llama_new_context_with_model: n_ctx_per_seq (65536) < n_ctx_train (131072) -- the full capacity of the model will not be utilized llama_kv_cache_init: kv_size = 65536, offload = 1, type_k = 'q4_0', type_v = 'q4_0', n_layer = 80, can_shift = 1 llama_kv_cache_init: ROCm0 KV buffer size = 2880.00 MiB llama_kv_cache_init: ROCm1 KV buffer size = 2880.00 MiB llama_new_context_with_model: KV self size = 5760.00 MiB, K (q4_0): 2880.00 MiB, V (q4_0): 2880.00 MiB llama_new_context_with_model: CPU output buffer size = 0.52 MiB llama_new_context_with_model: ROCm0 compute buffer size = 1088.45 MiB llama_new_context_with_model: ROCm1 compute buffer size = 208.00 MiB llama_new_context_with_model: ROCm_Host compute buffer size = 144.01 MiB llama_new_context_with_model: graph nodes = 2247 llama_new_context_with_model: graph splits = 5 (with bs=512), 4 (with bs=1) time=2025-02-12T14:28:24.861Z level=INFO source=server.go:597 msg="llama runner started in 42.31 seconds" [GIN] 2025/02/12 - 14:28:24 | 200 | 49.3998048s | 127.0.0.1 | POST "/api/generate" [GIN] 2025/02/12 - 14:29:01 | 200 | 16.1885007s | 127.0.0.1 | POST "/api/chat" llama_model_loader: loaded meta data with 36 key-value pairs and 724 tensors from T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 (version GGUF V3 (latest)) llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. llama_model_loader: - kv 0: general.architecture str = llama llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Llama 3.1 70B Instruct 2024 12 llama_model_loader: - kv 3: general.version str = 2024-12 llama_model_loader: - kv 4: general.finetune str = Instruct llama_model_loader: - kv 5: general.basename str = Llama-3.1 llama_model_loader: - kv 6: general.size_label str = 70B llama_model_loader: - kv 7: general.license str = llama3.1 llama_model_loader: - kv 8: general.base_model.count u32 = 1 llama_model_loader: - kv 9: general.base_model.0.name str = Llama 3.1 70B llama_model_loader: - kv 10: general.base_model.0.organization str = Meta Llama llama_model_loader: - kv 11: general.base_model.0.repo_url str = https://huggingface.co/meta-llama/Lla... llama_model_loader: - kv 12: general.tags arr[str,5] = ["facebook", "meta", "pytorch", "llam... llama_model_loader: - kv 13: general.languages arr[str,7] = ["fr", "it", "pt", "hi", "es", "th", ... llama_model_loader: - kv 14: llama.block_count u32 = 80 llama_model_loader: - kv 15: llama.context_length u32 = 131072 llama_model_loader: - kv 16: llama.embedding_length u32 = 8192 llama_model_loader: - kv 17: llama.feed_forward_length u32 = 28672 llama_model_loader: - kv 18: llama.attention.head_count u32 = 64 llama_model_loader: - kv 19: llama.attention.head_count_kv u32 = 8 llama_model_loader: - kv 20: llama.rope.freq_base f32 = 500000.000000 llama_model_loader: - kv 21: llama.attention.layer_norm_rms_epsilon f32 = 0.000010 llama_model_loader: - kv 22: llama.attention.key_length u32 = 128 llama_model_loader: - kv 23: llama.attention.value_length u32 = 128 llama_model_loader: - kv 24: general.file_type u32 = 10 llama_model_loader: - kv 25: llama.vocab_size u32 = 128256 llama_model_loader: - kv 26: llama.rope.dimension_count u32 = 128 llama_model_loader: - kv 27: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 28: tokenizer.ggml.pre str = llama-bpe llama_model_loader: - kv 29: tokenizer.ggml.tokens arr[str,128256] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 30: tokenizer.ggml.token_type arr[i32,128256] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 31: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "... llama_model_loader: - kv 32: tokenizer.ggml.bos_token_id u32 = 128000 llama_model_loader: - kv 33: tokenizer.ggml.eos_token_id u32 = 128009 llama_model_loader: - kv 34: tokenizer.chat_template str = {{- bos_token }}\n{%- if custom_tools ... llama_model_loader: - kv 35: general.quantization_version u32 = 2 llama_model_loader: - type f32: 162 tensors llama_model_loader: - type q2_K: 321 tensors llama_model_loader: - type q3_K: 160 tensors llama_model_loader: - type q5_K: 80 tensors llama_model_loader: - type q6_K: 1 tensors llm_load_vocab: special tokens cache size = 256 llm_load_vocab: token to piece cache size = 0.7999 MB llm_load_print_meta: format = GGUF V3 (latest) llm_load_print_meta: arch = llama llm_load_print_meta: vocab type = BPE llm_load_print_meta: n_vocab = 128256 llm_load_print_meta: n_merges = 280147 llm_load_print_meta: vocab_only = 1 llm_load_print_meta: model type = ?B llm_load_print_meta: model ftype = all F32 llm_load_print_meta: model params = 70.55 B llm_load_print_meta: model size = 24.56 GiB (2.99 BPW) llm_load_print_meta: general.name = Llama 3.1 70B Instruct 2024 12 llm_load_print_meta: BOS token = 128000 '<|begin_of_text|>' llm_load_print_meta: EOS token = 128009 '<|eot_id|>' llm_load_print_meta: EOT token = 128009 '<|eot_id|>' llm_load_print_meta: EOM token = 128008 '<|eom_id|>' llm_load_print_meta: LF token = 128 'Ä' llm_load_print_meta: EOG token = 128008 '<|eom_id|>' llm_load_print_meta: EOG token = 128009 '<|eot_id|>' llm_load_print_meta: max token length = 256 llama_model_load: vocab only - skipping tensors [GIN] 2025/02/12 - 14:29:12 | 200 | 9.5654491s | 127.0.0.1 | POST "/api/chat" [GIN] 2025/02/12 - 14:29:21 | 200 | 3.4621003s | 127.0.0.1 | POST "/api/chat" [GIN] 2025/02/12 - 14:29:28 | 200 | 4.4678599s | 127.0.0.1 | POST "/api/chat" [GIN] 2025/02/12 - 14:29:47 | 200 | 0s | 127.0.0.1 | HEAD "/" [GIN] 2025/02/12 - 14:29:47 | 200 | 23.6152ms | 127.0.0.1 | POST "/api/show" time=2025-02-12T14:29:54.457Z level=INFO source=sched.go:730 msg="new model will fit in available VRAM, loading" model=T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 library=rocm parallel=4 required="28.6 GiB" time=2025-02-12T14:29:55.232Z level=INFO source=server.go:100 msg="system memory" total="63.9 GiB" free="38.2 GiB" free_swap="26.1 GiB" time=2025-02-12T14:29:56.034Z level=INFO source=memory.go:356 msg="offload to rocm" layers.requested=-1 layers.model=81 layers.offload=81 layers.split=41,40 memory.available="[22.4 GiB 22.4 GiB]" memory.gpu_overhead="0 B" memory.required.full="28.6 GiB" memory.required.partial="28.6 GiB" memory.required.kv="640.0 MiB" memory.required.allocations="[14.7 GiB 13.9 GiB]" memory.weights.total="24.1 GiB" memory.weights.repeating="23.3 GiB" memory.weights.nonrepeating="822.0 MiB" memory.graph.full="1.1 GiB" memory.graph.partial="1.1 GiB" time=2025-02-12T14:29:56.035Z level=INFO source=server.go:185 msg="enabling flash attention" time=2025-02-12T14:29:56.039Z level=INFO source=server.go:381 msg="starting llama server" cmd="T:\\ai\\ollama\\ollama.exe runner --model T:\\ai\\models\\ollama\\blobs\\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 --ctx-size 8192 --batch-size 512 --n-gpu-layers 81 --threads 8 --flash-attn --kv-cache-type q4_0 --parallel 4 --tensor-split 41,40 --port 40130" time=2025-02-12T14:29:56.044Z level=INFO source=sched.go:449 msg="loaded runners" count=1 time=2025-02-12T14:29:56.044Z level=INFO source=server.go:558 msg="waiting for llama runner to start responding" time=2025-02-12T14:29:56.045Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server error" time=2025-02-12T14:29:56.080Z level=INFO source=runner.go:936 msg="starting go runner" time=2025-02-12T14:29:56.081Z level=INFO source=runner.go:937 msg=system info="CPU : LLAMAFILE = 1 | CPU : LLAMAFILE = 1 | cgo(clang)" threads=8 time=2025-02-12T14:29:56.082Z level=INFO source=runner.go:995 msg="Server listening on 127.0.0.1:40130" time=2025-02-12T14:29:56.296Z level=INFO source=server.go:592 msg="waiting for server to become available" status="llm server loading model" ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no ggml_cuda_init: found 2 ROCm devices: Device 0: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no Device 1: AMD Radeon RX 7900 XTX, compute capability 11.0, VMM: no load_backend: loaded ROCm backend from T:\ai\ollama\lib\ollama\rocm\ggml-hip.dll load_backend: loaded CPU backend from T:\ai\ollama\lib\ollama\ggml-cpu-haswell.dll llama_load_model_from_file: using device ROCm0 (AMD Radeon RX 7900 XTX) - 24194 MiB free llama_load_model_from_file: using device ROCm1 (AMD Radeon RX 7900 XTX) - 24411 MiB free llama_model_loader: loaded meta data with 36 key-value pairs and 724 tensors from T:\ai\models\ollama\blobs\sha256-35a6401f84b6c06d3d87140f6a437240cd02f65cc27216043911cda2bdde9137 (version GGUF V3 (latest)) llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. llama_model_loader: - kv 0: general.architecture str = llama llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Llama 3.1 70B Instruct 2024 12 llama_model_loader: - kv 3: general.version str = 2024-12 llama_model_loader: - kv 4: general.finetune str = Instruct llama_model_loader: - kv 5: general.basename str = Llama-3.1 llama_model_loader: - kv 6: general.size_label str = 70B llama_model_loader: - kv 7: general.license str = llama3.1 llama_model_loader: - kv 8: general.base_model.count u32 = 1 llama_model_loader: - kv 9: general.base_model.0.name str = Llama 3.1 70B llama_model_loader: - kv 10: general.base_model.0.organization str = Meta Llama llama_model_loader: - kv 11: general.base_model.0.repo_url str = https://huggingface.co/meta-llama/Lla... llama_model_loader: - kv 12: general.tags arr[str,5] = ["facebook", "meta", "pytorch", "llam... llama_model_loader: - kv 13: general.languages arr[str,7] = ["fr", "it", "pt", "hi", "es", "th", ... llama_model_loader: - kv 14: llama.block_count u32 = 80 llama_model_loader: - kv 15: llama.context_length u32 = 131072 llama_model_loader: - kv 16: llama.embedding_length u32 = 8192 llama_model_loader: - kv 17: llama.feed_forward_length u32 = 28672 llama_model_loader: - kv 18: llama.attention.head_count u32 = 64 llama_model_loader: - kv 19: llama.attention.head_count_kv u32 = 8 llama_model_loader: - kv 20: llama.rope.freq_base f32 = 500000.000000 llama_model_loader: - kv 21: llama.attention.layer_norm_rms_epsilon f32 = 0.000010 llama_model_loader: - kv 22: llama.attention.key_length u32 = 128 llama_model_loader: - kv 23: llama.attention.value_length u32 = 128 llama_model_loader: - kv 24: general.file_type u32 = 10 llama_model_loader: - kv 25: llama.vocab_size u32 = 128256 llama_model_loader: - kv 26: llama.rope.dimension_count u32 = 128 llama_model_loader: - kv 27: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 28: tokenizer.ggml.pre str = llama-bpe llama_model_loader: - kv 29: tokenizer.ggml.tokens arr[str,128256] = ["!", "\"", "#", "$", "%", "&", "'", ... llama_model_loader: - kv 30: tokenizer.ggml.token_type arr[i32,128256] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ... llama_model_loader: - kv 31: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "... llama_model_loader: - kv 32: tokenizer.ggml.bos_token_id u32 = 128000 llama_model_loader: - kv 33: tokenizer.ggml.eos_token_id u32 = 128009 llama_model_loader: - kv 34: tokenizer.chat_template str = {{- bos_token }}\n{%- if custom_tools ... llama_model_loader: - kv 35: general.quantization_version u32 = 2 llama_model_loader: - type f32: 162 tensors llama_model_loader: - type q2_K: 321 tensors llama_model_loader: - type q3_K: 160 tensors llama_model_loader: - type q5_K: 80 tensors llama_model_loader: - type q6_K: 1 tensors llm_load_vocab: special tokens cache size = 256 llm_load_vocab: token to piece cache size = 0.7999 MB llm_load_print_meta: format = GGUF V3 (latest) llm_load_print_meta: arch = llama llm_load_print_meta: vocab type = BPE llm_load_print_meta: n_vocab = 128256 llm_load_print_meta: n_merges = 280147 llm_load_print_meta: vocab_only = 0 llm_load_print_meta: n_ctx_train = 131072 llm_load_print_meta: n_embd = 8192 llm_load_print_meta: n_layer = 80 llm_load_print_meta: n_head = 64 llm_load_print_meta: n_head_kv = 8 llm_load_print_meta: n_rot = 128 llm_load_print_meta: n_swa = 0 llm_load_print_meta: n_embd_head_k = 128 llm_load_print_meta: n_embd_head_v = 128 llm_load_print_meta: n_gqa = 8 llm_load_print_meta: n_embd_k_gqa = 1024 llm_load_print_meta: n_embd_v_gqa = 1024 llm_load_print_meta: f_norm_eps = 0.0e+00 llm_load_print_meta: f_norm_rms_eps = 1.0e-05 llm_load_print_meta: f_clamp_kqv = 0.0e+00 llm_load_print_meta: f_max_alibi_bias = 0.0e+00 llm_load_print_meta: f_logit_scale = 0.0e+00 llm_load_print_meta: n_ff = 28672 llm_load_print_meta: n_expert = 0 llm_load_print_meta: n_expert_used = 0 llm_load_print_meta: causal attn = 1 llm_load_print_meta: pooling type = 0 llm_load_print_meta: rope type = 0 llm_load_print_meta: rope scaling = linear llm_load_print_meta: freq_base_train = 500000.0 llm_load_print_meta: freq_scale_train = 1 llm_load_print_meta: n_ctx_orig_yarn = 131072 llm_load_print_meta: rope_finetuned = unknown llm_load_print_meta: ssm_d_conv = 0 llm_load_print_meta: ssm_d_inner = 0 llm_load_print_meta: ssm_d_state = 0 llm_load_print_meta: ssm_dt_rank = 0 llm_load_print_meta: ssm_dt_b_c_rms = 0 llm_load_print_meta: model type = 70B llm_load_print_meta: model ftype = Q2_K - Medium llm_load_print_meta: model params = 70.55 B llm_load_print_meta: model size = 24.56 GiB (2.99 BPW) llm_load_print_meta: general.name = Llama 3.1 70B Instruct 2024 12 llm_load_print_meta: BOS token = 128000 '<|begin_of_text|>' llm_load_print_meta: EOS token = 128009 '<|eot_id|>' llm_load_print_meta: EOT token = 128009 '<|eot_id|>' llm_load_print_meta: EOM token = 128008 '<|eom_id|>' llm_load_print_meta: LF token = 128 'Ä' llm_load_print_meta: EOG token = 128008 '<|eom_id|>' llm_load_print_meta: EOG token = 128009 '<|eot_id|>' llm_load_print_meta: max token length = 256 llm_load_tensors: offloading 80 repeating layers to GPU llm_load_tensors: offloading output layer to GPU llm_load_tensors: offloaded 81/81 layers to GPU llm_load_tensors: ROCm0 model buffer size = 12297.44 MiB llm_load_tensors: ROCm1 model buffer size = 12519.55 MiB llm_load_tensors: CPU_Mapped model buffer size = 328.78 MiB llama_new_context_with_model: n_seq_max = 4 llama_new_context_with_model: n_ctx = 8192 llama_new_context_with_model: n_ctx_per_seq = 2048 llama_new_context_with_model: n_batch = 2048 llama_new_context_with_model: n_ubatch = 512 llama_new_context_with_model: flash_attn = 1 llama_new_context_with_model: freq_base = 500000.0 llama_new_context_with_model: freq_scale = 1 llama_new_context_with_model: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized llama_kv_cache_init: kv_size = 8192, offload = 1, type_k = 'q4_0', type_v = 'q4_0', n_layer = 80, can_shift = 1 llama_kv_cache_init: ROCm0 KV buffer size = 369.00 MiB llama_kv_cache_init: ROCm1 KV buffer size = 351.00 MiB llama_new_context_with_model: KV self size = 720.00 MiB, K (q4_0): 360.00 MiB, V (q4_0): 360.00 MiB llama_new_context_with_model: ROCm_Host output buffer size = 2.08 MiB llama_new_context_with_model: pipeline parallelism enabled (n_copies=4) llama_new_context_with_model: ROCm0 compute buffer size = 296.01 MiB llama_new_context_with_model: ROCm1 compute buffer size = 362.52 MiB llama_new_context_with_model: ROCm_Host compute buffer size = 80.02 MiB llama_new_context_with_model: graph nodes = 2247 llama_new_context_with_model: graph splits = 3 time=2025-02-12T14:30:35.095Z level=INFO source=server.go:597 msg="llama runner started in 39.05 seconds" [GIN] 2025/02/12 - 14:30:35 | 200 | 47.7650191s | 127.0.0.1 | POST "/api/generate" [GIN] 2025/02/12 - 14:30:44 | 200 | 2.87911s | 127.0.0.1 | POST "/api/chat" [GIN] 2025/02/12 - 14:31:30 | 200 | 0s | 127.0.0.1 | GET "/api/version" (base) PS C:\WINDOWS\system32> ``` ### OS Windows ### GPU AMD ### CPU AMD ### Ollama version 0.5.9
GiteaMirror added the bug label 2026-04-22 12:18:18 -05:00
GiteaMirror commented 2026-04-22 12:18:21 -05:00
Author
Owner
Copy Link

@TheSeraph commented on GitHub (Feb 12, 2025):

I tried this with 0.5.8 and it had a very similar output. Just for kicks, I deleted eveyrthing and reinstall from scratch. For whatever reason the new stack doesn't seem to interact well. I've got flash attention enabled with Q4 KV cache quantization. Not sure if that's an issue or not

<!-- gh-comment-id:2653935644 --> @TheSeraph commented on GitHub (Feb 12, 2025): I tried this with 0.5.8 and it had a very similar output. Just for kicks, I deleted eveyrthing and reinstall from scratch. For whatever reason the new stack doesn't seem to interact well. I've got flash attention enabled with Q4 KV cache quantization. Not sure if that's an issue or not
GiteaMirror commented 2026-04-22 12:18:22 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Feb 12, 2025):

A few more questions to help troubleshooting:

  • Are you able to try 0.5.7?
  • Have you tried any other models? Do they exhibit the same behavior?
  • Have you tried turning off flash attention and KV cache quantization?
<!-- gh-comment-id:2654348348 --> @mxyng commented on GitHub (Feb 12, 2025): A few more questions to help troubleshooting: * Are you able to try 0.5.7? * Have you tried any other models? Do they exhibit the same behavior? * Have you tried turning off flash attention and KV cache quantization?
GiteaMirror commented 2026-04-22 12:18:22 -05:00
Author
Owner
Copy Link

@TheSeraph commented on GitHub (Feb 12, 2025):

So, everything seems to work fine with 0.5.7 as I reverted with no issues. I installed tha amd64 package in a seperate folder, turning off flash attention, and setting KV cahce to F16

Image

The behaviour improves, but it's still gibberish at least with llama3.3:70b-instruct-q2_K

Image

Deepseek-r1:8b did not seem to exhibit the same insanity on first glance

Image

Similar results were acheived with mistral-nemo both at the standard 2048 context window and my custom 64k context window

I tried to go back to my settings (FA True, KV q4) and those other models seemed to work as well, so perhaps this is just a llama3.3 problem? I tried llama3.1:8b-instruct-fp16 and found no issues. I tried llama3.3:70b-instruct-q3_K_S as well, and that also did the gibberish thing.

<!-- gh-comment-id:2655002390 --> @TheSeraph commented on GitHub (Feb 12, 2025): So, everything seems to work fine with 0.5.7 as I reverted with no issues. I installed tha amd64 package in a seperate folder, turning off flash attention, and setting KV cahce to F16 ![Image](https://github.com/user-attachments/assets/6956b77e-d2ad-46c0-8abb-7b1526c33276) The behaviour improves, but it's still gibberish at least with llama3.3:70b-instruct-q2_K ![Image](https://github.com/user-attachments/assets/d6f7b6df-b6f4-411e-8953-93d9e37eec0f) Deepseek-r1:8b did not seem to exhibit the same insanity on first glance ![Image](https://github.com/user-attachments/assets/607b1b80-b244-4ee9-96e9-10c7a47a6de8) Similar results were acheived with mistral-nemo both at the standard 2048 context window and my custom 64k context window I tried to go back to my settings (FA True, KV q4) and those other models seemed to work as well, so perhaps this is just a llama3.3 problem? I tried llama3.1:8b-instruct-fp16 and found no issues. I tried llama3.3:70b-instruct-q3_K_S as well, and that also did the gibberish thing.
GiteaMirror commented 2026-04-22 12:18:23 -05:00
Author
Owner
Copy Link

@TheSeraph commented on GitHub (Feb 12, 2025):

For fun, I thought perhaps I should try a larger model, so I tried deepseek-r1:32b with 64k context and it also spits out total crap. I tried the same but setup as standard (2048 ctx window), which seems to run on just one GPU and it works. Is there perhaps an issue with larger models, or perhaps an issue when split across two AMD GPUs? I tried Command-R which is bigger than 1 GPU and had garbage out as well, seems there's some issues with splitting across multiple GPUs in 0.5.9

<!-- gh-comment-id:2655023869 --> @TheSeraph commented on GitHub (Feb 12, 2025): For fun, I thought perhaps I should try a larger model, so I tried deepseek-r1:32b with 64k context and it also spits out total crap. I tried the same but setup as standard (2048 ctx window), which seems to run on just one GPU and it works. Is there perhaps an issue with larger models, or perhaps an issue when split across two AMD GPUs? I tried Command-R which is bigger than 1 GPU and had garbage out as well, seems there's some issues with splitting across multiple GPUs in 0.5.9
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#31649
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?