github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 13:19:40 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #5358] LLM Compiler Models #3355

New Issue
Closed
opened 2026-04-12 13:58:13 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-12 13:58:13 -05:00
Owner
Copy Link

Originally created by @pmatos on GitHub (Jun 28, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/5358

How can I use the newly release models from meta with Ollama?

https://huggingface.co/collections/facebook/llm-compiler-667c5b05557fe99a9edd25cb

Thanks.

Originally created by @pmatos on GitHub (Jun 28, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/5358 How can I use the newly release models from meta with Ollama? https://huggingface.co/collections/facebook/llm-compiler-667c5b05557fe99a9edd25cb Thanks.
GiteaMirror added the model label 2026-04-12 13:58:13 -05:00
GiteaMirror commented 2026-04-12 13:58:14 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 30, 2024):

Download the model. It's a gated model so you have to apply for and be granted access, and you will need to generate a token.

huggingface-cli download --token TOKEN  --local-dir . facebook/llm-compiler-7b-ftd

Convert to GGUF format. I find using a dockerized llama.cpp is easiest, if you don't have docker you will need to install llama.cpp.

docker run -it -v .:/models ghcr.io/ggerganov/llama.cpp:full-cuda -c --outtype f16 /models

Quantize it as required:

docker run -it --gpus all -v .:/models ghcr.io/ggerganov/llama.cpp:full-cuda -q /models/ggml-model-f16.gguf Q4_K_S

Create a model file:

echo FROM ggml-model-Q4_K_S.gguf > Modelfile

Create the model:

ollama create llm-compiler:7b-ftd-Q4_K_S -f Modelfile

Try it out. Extract the assembly example from the test file, wrap it in a prompt and send it to the model:

sed -ne '/asm =/,/"""/p' llm_compiler_demo.py | grep -v '"""' > asm
curl -s http://localhost:11434/api/generate -d '{"model":"llm-compiler:7b-ftd-Q4_K_S","prompt": '"$((echo -e '[INST] Disassemble this code to LLVM-IR:\n\n<code>' ; cat asm ; echo '</code> [/INST]') | jq -sR)"',"format":"","options":{},"stream": false, "raw": true}'

The response looks like this:

This will produce code:

<code>; ModuleID = '<stdin>'
source_filename = "-"
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-none-linux-gnu"

; Function Attrs: minsize nofree norecurse nounwind optsize memory(inaccessiblemem: readwrite) uwtable
define dso_local i32 @add_two(i32 noundef %0, i32 noundef %1) local_unnamed_addr #0 {
  %3 = alloca i32, align 4
  %4 = alloca i32, align 4
  store volatile i32 %0, ptr %3, align 4, !tbaa !6
  store volatile i32 %1, ptr %4, align 4, !tbaa !6
  %.0..0..0..0.2 = load volatile i32, ptr %3, align 4, !tbaa !6
  %.0..0..0..0.1 = load volatile i32, ptr %4, align 4, !tbaa !6
  %5 = add nsw i32 %.0..0..0..0.1, %.0..0..0..0.2
  ret i32 %5
}

attributes #0 = { minsize nofree norecurse nounwind optsize memory(inaccessiblemem: readwrite) uwtable "frame-pointer"="non-leaf" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+fp-armv8,+neon,+outline-atomics,+v8a,-fmv" }

!llvm.module.flags = !{!0, !1, !2, !3, !4}
!llvm.ident = !{!5}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{i32 8, !"PIC Level", i32 2}
!2 = !{i32 7, !"PIE Level", i32 2}
!3 = !{i32 7, !"uwtable", i32 2}
!4 = !{i32 7, !"frame-pointer", i32 1}
!5 = !{!"clang version 17.0.6 (git@github.com:fairinternal/CodeGen.git b05db9bbf7a92019267416c1bb9996fe6134e3f1)"}
!6 = !{!7, !7, i64 0}
!7 = !{!"int", !8, i64 0}
!8 = !{!"omnipotent char", !9, i64 0}
!9 = !{!"Simple C/C++ TBAA"}
</code>

How correct this after quant is TBD.

<!-- gh-comment-id:2198407660 --> @rick-github commented on GitHub (Jun 30, 2024): Download the model. It's a gated model so you have to apply for and be granted access, and you will need to generate a token. ``` huggingface-cli download --token TOKEN --local-dir . facebook/llm-compiler-7b-ftd ``` Convert to GGUF format. I find using a dockerized llama.cpp is easiest, if you don't have docker you will need to install llama.cpp. ``` docker run -it -v .:/models ghcr.io/ggerganov/llama.cpp:full-cuda -c --outtype f16 /models ``` Quantize it as required: ``` docker run -it --gpus all -v .:/models ghcr.io/ggerganov/llama.cpp:full-cuda -q /models/ggml-model-f16.gguf Q4_K_S ``` Create a model file: ``` echo FROM ggml-model-Q4_K_S.gguf > Modelfile ``` Create the model: ``` ollama create llm-compiler:7b-ftd-Q4_K_S -f Modelfile ``` Try it out. Extract the assembly example from the test file, wrap it in a prompt and send it to the model: ``` sed -ne '/asm =/,/"""/p' llm_compiler_demo.py | grep -v '"""' > asm curl -s http://localhost:11434/api/generate -d '{"model":"llm-compiler:7b-ftd-Q4_K_S","prompt": '"$((echo -e '[INST] Disassemble this code to LLVM-IR:\n\n<code>' ; cat asm ; echo '</code> [/INST]') | jq -sR)"',"format":"","options":{},"stream": false, "raw": true}' ``` The response looks like this: ``` This will produce code: <code>; ModuleID = '<stdin>' source_filename = "-" target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128" target triple = "aarch64-none-linux-gnu" ; Function Attrs: minsize nofree norecurse nounwind optsize memory(inaccessiblemem: readwrite) uwtable define dso_local i32 @add_two(i32 noundef %0, i32 noundef %1) local_unnamed_addr #0 { %3 = alloca i32, align 4 %4 = alloca i32, align 4 store volatile i32 %0, ptr %3, align 4, !tbaa !6 store volatile i32 %1, ptr %4, align 4, !tbaa !6 %.0..0..0..0.2 = load volatile i32, ptr %3, align 4, !tbaa !6 %.0..0..0..0.1 = load volatile i32, ptr %4, align 4, !tbaa !6 %5 = add nsw i32 %.0..0..0..0.1, %.0..0..0..0.2 ret i32 %5 } attributes #0 = { minsize nofree norecurse nounwind optsize memory(inaccessiblemem: readwrite) uwtable "frame-pointer"="non-leaf" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+fp-armv8,+neon,+outline-atomics,+v8a,-fmv" } !llvm.module.flags = !{!0, !1, !2, !3, !4} !llvm.ident = !{!5} !0 = !{i32 1, !"wchar_size", i32 4} !1 = !{i32 8, !"PIC Level", i32 2} !2 = !{i32 7, !"PIE Level", i32 2} !3 = !{i32 7, !"uwtable", i32 2} !4 = !{i32 7, !"frame-pointer", i32 1} !5 = !{!"clang version 17.0.6 (git@github.com:fairinternal/CodeGen.git b05db9bbf7a92019267416c1bb9996fe6134e3f1)"} !6 = !{!7, !7, i64 0} !7 = !{!"int", !8, i64 0} !8 = !{!"omnipotent char", !9, i64 0} !9 = !{!"Simple C/C++ TBAA"} </code> ``` How correct this after quant is TBD.
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 model
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#3355
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?