github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[PR #10634] Add Huggingface model card README.md (YAML) to GGUF converter #39175

New Issue
Open
opened 2026-04-22 23:49:36 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-22 23:49:36 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/10634
Author: @mrutkows
Created: 5/9/2025
Status: 🔄 Open

Base: main ← Head: convert-hf-metadata2

📝 Commits (1)

  • 825619c Add Huggingface model card README.md (YAML) to GGUF convert

📊 Changes

15 files changed (+469 additions, -3 deletions)

View changed files

📝 convert/convert.go (+23 -3)
➕ convert/convert_modelcard.go (+148 -0)
➕ convert/convert_modelcard_test.go (+96 -0)
➕ convert/testdata/modelcard/README md.Llama-4-Maverick-17B-128E-Instruct (+102 -0)
➕ convert/testdata/modelcard/README.md.OpenHermes-2-Mistral-7B (+23 -0)
➕ convert/testdata/modelcard/README.md.Phi-3.5-mini-instruct (+18 -0)
➕ convert/testdata/modelcard/README.md.Qwen2.5-7B-Instruct (+14 -0)
➕ convert/testdata/modelcard/README.md.bad_delim (+9 -0)
➕ convert/testdata/modelcard/README.md.empty_array (+6 -0)
➕ convert/testdata/modelcard/README.md.missing_delim (+5 -0)
➕ convert/testdata/modelcard/README.md.tiny-LlamaForCausalLM-3.2 (+9 -0)
📝 fs/ggml/ggml.go (+6 -0)
📝 go.mod (+1 -0)
📝 go.sum (+2 -0)
📝 parser/parser.go (+7 -0)

📄 Description

Huggingface has standardized a set of YAML (tags) that describe such things as license information, model provenance and organizational ownership that is supported by the GGUF format and can be added as general.xx entries in the KV header values (see https://huggingface.co/docs/hub/en/model-cards#editing-the-yaml-section-of-the-readmemd-file and https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/gguf/constants.py respectively).

This PR seeks to automate that inclusion of that data as part of the model conversion (i.e., create) process since this information is stored at the top of each Huggingface model's README.md (even templated on model repo. creation) and is fairly easily extracted, parsed and mapped.

Please note that only the information that is relevant to the model itself, from a larger set HF supports, is brought over and aligns with what has been done in other converters.

Testing

See actual README.md YAML snippets for the following models which provided a good cross-section of use of the YAML schema:
- Llama-4-Maverick-17B-128E-Instruct
- OpenHermes-2-Mistral-7B
- Phi-3.5-mini-instruct
- Qwen2.5-7B-Instruct
- tiny-LlamaForCausalLM-3.2

As well as unit tests for incorrect YAML, etc.:
- bad_delim
- missing_delim
- empty_array

Post "create" (convert) testing of actual HF models:

  • Included new unit/functional tests for a variety of popular models with a good representation of different YAML metadata
  • Tested "create" API on the following models and independently ran the GGUFs on llama.cpp for additional validation:
    • granite-3.2-2b-instruct, granite-3.1-3b-a800m-instruct (with PRs waiting for review)
    • Phi-3.5-mini-instruct
    • Qwen2.5-7B-Instruct
    • tiny-LlamaForCausalLM-3.2

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/ollama/ollama/pull/10634 **Author:** [@mrutkows](https://github.com/mrutkows) **Created:** 5/9/2025 **Status:** 🔄 Open **Base:** `main` ← **Head:** `convert-hf-metadata2` --- ### 📝 Commits (1) - [`825619c`](https://github.com/ollama/ollama/commit/825619c38e27920c63841c108eb682244bbbbf09) Add Huggingface model card README.md (YAML) to GGUF convert ### 📊 Changes **15 files changed** (+469 additions, -3 deletions) <details> <summary>View changed files</summary> 📝 `convert/convert.go` (+23 -3) ➕ `convert/convert_modelcard.go` (+148 -0) ➕ `convert/convert_modelcard_test.go` (+96 -0) ➕ `convert/testdata/modelcard/README md.Llama-4-Maverick-17B-128E-Instruct` (+102 -0) ➕ `convert/testdata/modelcard/README.md.OpenHermes-2-Mistral-7B` (+23 -0) ➕ `convert/testdata/modelcard/README.md.Phi-3.5-mini-instruct` (+18 -0) ➕ `convert/testdata/modelcard/README.md.Qwen2.5-7B-Instruct` (+14 -0) ➕ `convert/testdata/modelcard/README.md.bad_delim` (+9 -0) ➕ `convert/testdata/modelcard/README.md.empty_array` (+6 -0) ➕ `convert/testdata/modelcard/README.md.missing_delim` (+5 -0) ➕ `convert/testdata/modelcard/README.md.tiny-LlamaForCausalLM-3.2` (+9 -0) 📝 `fs/ggml/ggml.go` (+6 -0) 📝 `go.mod` (+1 -0) 📝 `go.sum` (+2 -0) 📝 `parser/parser.go` (+7 -0) </details> ### 📄 Description Huggingface has standardized a set of YAML (tags) that describe such things as license information, model provenance and organizational ownership that is supported by the GGUF format and can be added as general.xx entries in the KV header values (see https://huggingface.co/docs/hub/en/model-cards#editing-the-yaml-section-of-the-readmemd-file and https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/gguf/constants.py respectively). This PR seeks to automate that inclusion of that data as part of the model conversion (i.e., create) process since this information is stored at the top of each Huggingface model's README.md (even templated on model repo. creation) and is fairly easily extracted, parsed and mapped. Please note that only the information that is relevant to the model itself, from a larger set HF supports, is brought over and aligns with what has been done in other converters. ### Testing See actual README.md YAML snippets for the following models which provided a good cross-section of use of the YAML schema: - Llama-4-Maverick-17B-128E-Instruct - OpenHermes-2-Mistral-7B - Phi-3.5-mini-instruct - Qwen2.5-7B-Instruct - tiny-LlamaForCausalLM-3.2 As well as unit tests for incorrect YAML, etc.: - bad_delim - missing_delim - empty_array #### Post "create" (convert) testing of actual HF models: - Included new unit/functional tests for a variety of popular models with a good representation of different YAML metadata - Tested "create" API on the following models and independently ran the GGUFs on llama.cpp for additional validation: - granite-3.2-2b-instruct, granite-3.1-3b-a800m-instruct (with PRs waiting for review) - Phi-3.5-mini-instruct - Qwen2.5-7B-Instruct - tiny-LlamaForCausalLM-3.2 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
GiteaMirror added the pull-request label 2026-04-22 23:49:36 -05:00
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 pull-request
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#39175
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?