github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #7223] How to add support for RWKV? #51095

New Issue
Closed
opened 2026-04-28 18:19:59 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-28 18:19:59 -05:00
Owner
Copy Link

Originally created by @MollySophia on GitHub (Oct 16, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7223

Hi! I would like to try to make RWKV v6 models working with ollama.
llama.cpp has it supported already.

  • Currently ollama fails to load the model due to a bug in llama.cpp. Here's the fix PR: https://github.com/ggerganov/llama.cpp/pull/9907
  • Another issue is the chat template. I wonder how should a chat template be added for a new model? Specifically, how does ollama decide which template to use when loading a modelfile?

Thanks!

Originally created by @MollySophia on GitHub (Oct 16, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7223 Hi! I would like to try to make RWKV v6 models working with ollama. llama.cpp has it supported already. - Currently ollama fails to load the model due to a bug in llama.cpp. Here's the fix PR: https://github.com/ggerganov/llama.cpp/pull/9907 - Another issue is the chat template. I wonder how should a chat template be added for a new model? Specifically, how does ollama decide which template to use when loading a modelfile? Thanks!
GiteaMirror added the model label 2026-04-28 18:19:59 -05:00
GiteaMirror commented 2026-04-28 18:20:01 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 16, 2024):

ollama looks for the field tokenizer.chat_template in a GGUF file and then searches for a match in template/index.json. If it finds a match it will use the ollama template from the template directory. It doesn't look like RWKV publishes an official GGUF file so you are dependent on third parties. I had a look at Lyte/RWKV-6-World-3B-v2.1-GGUF and the files don't include a chat_template, so ollama will fall back to a default of {{ .Prompt }}. However, looking at the RWKV's example usage the template looks fairly simple:

TEMPLATE="""{{ if .System }}Instruction: {{ .System }}

Input: {{ .Prompt }}

Response:
{{- else }}User: Hi

Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.

User: {{ .Prompt }}

Assistant:{{ end }}"""
<!-- gh-comment-id:2416439672 --> @rick-github commented on GitHub (Oct 16, 2024): ollama looks for the field `tokenizer.chat_template` in a GGUF file and then searches for a match in [`template/index.json`](https://github.com/ollama/ollama/blob/main/template/index.json). If it finds a match it will use the ollama template from the [`template`](https://github.com/ollama/ollama/tree/main/template) directory. It doesn't look like [RWKV](https://huggingface.co/collections/RWKV/rwkv-v6-669cb221fe9496b3c693c8e9) publishes an official GGUF file so you are dependent on third parties. I had a look at [Lyte/RWKV-6-World-3B-v2.1-GGUF](https://huggingface.co/Lyte/RWKV-6-World-3B-v2.1-GGUF/tree/main) and the files don't include a chat_template, so ollama will fall back to a default of `{{ .Prompt }}`. However, looking at the RWKV's [example usage](https://huggingface.co/RWKV/v6-Finch-14B-HF/blob/main/README.md) the template looks fairly simple: ```modelfile TEMPLATE="""{{ if .System }}Instruction: {{ .System }} Input: {{ .Prompt }} Response: {{- else }}User: Hi Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it. User: {{ .Prompt }} Assistant:{{ end }}""" ```
GiteaMirror commented 2026-04-28 18:20:02 -05:00
Author
Owner
Copy Link

@MollySophia commented on GitHub (Oct 16, 2024):

ollama looks for the field tokenizer.chat_template in a GGUF file and then searches for a match in template/index.json. If it finds a match it will use the ollama template from the template directory. It doesn't look like RWKV publishes an official GGUF file so you are dependent on third parties. I had a look at Lyte/RWKV-6-World-3B-v2.1-GGUF and the files don't include a chat_template, so ollama will fall back to a default of {{ .Prompt }}.

The llama.cpp convert_hf_to_gguf.py tries to parse tokenizer_config.json to get the chat_template field. However, RWKV hf models' tokenizer_config doesn't specify that. RWKV uses an unique template anyway.

However, looking at the RWKV's example usage the template looks fairly simple:

Yes. It's really simple.

<!-- gh-comment-id:2416566443 --> @MollySophia commented on GitHub (Oct 16, 2024): > ollama looks for the field `tokenizer.chat_template` in a GGUF file and then searches for a match in [`template/index.json`](https://github.com/ollama/ollama/blob/main/template/index.json). If it finds a match it will use the ollama template from the [`template`](https://github.com/ollama/ollama/tree/main/template) directory. It doesn't look like [RWKV](https://huggingface.co/collections/RWKV/rwkv-v6-669cb221fe9496b3c693c8e9) publishes an official GGUF file so you are dependent on third parties. I had a look at [Lyte/RWKV-6-World-3B-v2.1-GGUF](https://huggingface.co/Lyte/RWKV-6-World-3B-v2.1-GGUF/tree/main) and the files don't include a chat_template, so ollama will fall back to a default of `{{ .Prompt }}`. The llama.cpp `convert_hf_to_gguf.py` tries to parse `tokenizer_config.json` to get the chat_template field. However, RWKV hf models' tokenizer_config doesn't specify that. RWKV uses an unique template anyway. > However, looking at the RWKV's [example usage](https://huggingface.co/RWKV/v6-Finch-14B-HF/blob/main/README.md) the template looks fairly simple: Yes. It's really simple.
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#51095
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?