初始化项目，由ModelHub XC社区提供模型

Model: javierdejesusda/yuholens-14b-GGUF Source: Original Platform
2026-06-21 06:22:17 +08:00
commit f427535be3
7 changed files with 199 additions and 0 deletions
--- a/.gitattributes
+++ b/.gitattributes
@@ -0,0 +1,40 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+yuholens-14b-Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+yuholens-14b-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+yuholens-14b-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+yuholens-14b-Q6_K.gguf filter=lfs diff=lfs merge=lfs -text
+yuholens-14b-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
--- a/README.md
+++ b/README.md
@@ -0,0 +1,144 @@
+---
+license: other
+license_name: tongyi-qianwen
+license_link: https://huggingface.co/Qwen/Qwen-14B/blob/main/LICENSE
+language:
+  - ja
+  - en
+base_model: javierdejesusda/yuholens-14b
+tags:
+  - gguf
+  - llama-cpp
+  - quantized
+  - japanese-finance
+  - yuho
+  - edinet
+  - qwen
+pipeline_tag: text-generation
+---
+
+# YuhoLens-14B GGUF
+
+GGUF-quantized release of [`javierdejesusda/yuholens-14b`](https://huggingface.co/javierdejesusda/yuholens-14b) for offline inference via [`llama.cpp`](https://github.com/ggerganov/llama.cpp). The BF16 source is a full-parameter SFT of [`pfnet/nekomata-14b-pfn-qfin`](https://huggingface.co/pfnet/nekomata-14b-pfn-qfin) (Qwen1, 14B, Japanese-finance CPT) on teacher-bootstrapped English investor memos derived from [`SakanaAI/EDINET-Bench`](https://huggingface.co/datasets/SakanaAI/EDINET-Bench).
+
+## Files
+
+| File | Quant | Size | Bits/weight | Recommended hardware |
+|------|-------|-----:|:-----------:|----------------------|
+| `yuholens-14b-Q3_K_M.gguf` | Q3_K_M | 7.18 GB |  4.35 | 8 GB GPU (RTX 4070 Laptop, 3060 Ti) — fits with `--ctx-size 2048` |
+| `yuholens-14b-Q4_K_M.gguf` | Q4_K_M | 8.81 GB | ~5.0 | 12-16 GB GPU (RTX 4060 Ti 16 GB, RTX 3080) |
+| `yuholens-14b-Q5_K_M.gguf` | Q5_K_M | 9.94 GB | ~5.7 | 16 GB GPU |
+| `yuholens-14b-Q6_K.gguf`   | Q6_K   | 11.46 GB | ~6.6 | 16-24 GB GPU |
+| `yuholens-14b-Q8_0.gguf`   | Q8_0   | 14.03 GB | 8.5 | 24 GB GPU or CPU offload |
+
+Sizes are the actual on-disk byte counts; quoted GB values use 1024³.
+
+## Smoke test
+
+The Q3_K_M quant has been smoke-tested on an NVIDIA RTX 4070 Laptop (8 GB VRAM, compute 8.9) using `llama-completion` from llama.cpp build `b8966`.
+
+- Prompt: ChatML-wrapped Yuho fixture with the YuhoLens system prompt (see *Prompt format* below).
+- Settings: `--n-gpu-layers 99 -c 2048 --temp 0.1 --top-p 0.9 --repeat-penalty 1.15`.
+- Throughput: prompt eval **65.5 tok/s**, generation **12.2 tok/s**.
+- VRAM occupancy: ~7.0 GB model + ~1.6 GB context + ~0.3 GB compute, fits within the 8.0 GB card.
+- Output: coherent English investor memo with section headings (*Executive summary*, *Going-concern assessment*, *Accrual quality*, ...) following the SFT teacher template.
+
+Q4_K_M and larger quants exceed 8 GB VRAM and require either a 12 GB+ GPU or partial CPU offload (`--n-gpu-layers` < 99).
+
+## Quickstart
+
+```bash
+# Q3_K_M, fits 8 GB GPU
+llama-completion \
+  -m yuholens-14b-Q3_K_M.gguf \
+  -f your_prompt.txt \
+  --n-gpu-layers 99 \
+  -c 2048 \
+  --temp 0.1 --top-p 0.9 --repeat-penalty 1.15 \
+  -n 512 \
+  --no-display-prompt
+```
+
+For larger quants on a smaller GPU, drop `--n-gpu-layers` to a value that fits VRAM (e.g. `--n-gpu-layers 30` for Q4_K_M on 8 GB).
+
+## Prompt format
+
+YuhoLens-14B was fine-tuned on Qwen1 ChatML with a fixed system prompt. **Raw Japanese Yuho text without the wrapper produces Japanese continuations** instead of English memos. Wrap inputs as:
+
+```
+<|im_start|>system
+{SYSTEM_PROMPT}<|im_end|>
+<|im_start|>user
+Company metadata (JSON):
+{...}
+
+Balance sheet (JSON):
+{...}
+
+P&L (JSON):
+{...}
+
+Cash flow (JSON):
+{...}
+
+Japanese annual-report text (truncated at ~20K chars):
+<<<
+{Japanese Yuho text}
+>>>
+
+Produce the two-page English investor memo now.<|im_end|>
+<|im_start|>assistant
+```
+
+The full system prompt is published in the BF16 model card and in `src/yuholens/training/teacher.py` of the GitHub repo. A pre-formatted smoke fixture is included in the GitHub repo at `data/sample/smoke_prompt_chatml.txt`.
+
+## Build provenance
+
+Source checkpoint: `output/yuholens-14b-sft/checkpoint-212` (28.3 GB BF16 safetensors), tagged at git commit `f903174`.
+
+Built with:
+
+- `llama.cpp` commit `b8966` (`7b8443ac7`), Windows CUDA 12.4 prebuilt binaries.
+- Convert script `convert_hf_to_gguf.py` patched locally to fall back to an inline GPT-2 byte-to-unicode mapping when transformers ≥ 5.x removes `bytes_to_unicode`. See the GitHub repo for the patch.
+- `llama-quantize` invoked with `--override-kv qwen.attention.layer_norm_rms_epsilon=float:0.000001` because Qwen1 uses RMSNorm internally but its HF config exposes the field as `layer_norm_epsilon`; this override is now baked into `scripts/build_gguf.sh`.
+
+Reproduce locally:
+
+```bash
+LLAMACPP_REPO=/path/to/llama.cpp \
+LLAMACPP_BIN=/path/to/llama.cpp/build/bin \
+  scripts/build_gguf.sh output/yuholens-14b-sft/checkpoint-212 data/eval/gguf
+```
+
+## Limitations
+
+- **Output language asymmetry.** The model emits English memos and expects Japanese Yuho input. Japanese output is unsupported by training distribution.
+- **Citation accuracy unaudited.** The model produces inline `(ref: '<span>' p.N)` citations, but verbatim-correctness against the underlying Yuho text has not been audited.
+- **Quantization quality bands.** Q3_K_M sacrifices some fidelity for the 8 GB VRAM target. Prefer Q4_K_M or higher when memory permits.
+- **Domain.** Trained on Yuho 有価証券報告書 only. Earnings transcripts, 決算短信, and non-Japanese filings are out-of-scope.
+- **Not financial advice.** Outputs are model-generated and may contain factual errors. Verify every material claim against the source.
+
+## License
+
+Released under the Tongyi Qianwen License inherited from the Qwen1 base via `pfnet/nekomata-14b-pfn-qfin`. See the linked license for terms. Wrapper code in the GitHub repo (LangGraph pipeline, training and evaluation scripts, prompt modules) is MIT.
+
+## Citation
+
+```bibtex
+@misc{dejesus2026yuholens,
+  author       = {De Jesus, Javier},
+  title        = {YuhoLens-14B: A Japanese-Finance Fine-Tune for
+                  Span-Grounded Investor Memo Generation},
+  year         = {2026},
+  howpublished = {Hugging Face model repository},
+  url          = {https://huggingface.co/javierdejesusda/yuholens-14b},
+  note         = {DOI: TBD}
+}
+```
+
+## Links
+
+- BF16 reference: https://huggingface.co/javierdejesusda/yuholens-14b
+- GitHub: https://github.com/javierdejesusda/YuhoLens
+- Base model: https://huggingface.co/pfnet/nekomata-14b-pfn-qfin
+- Training data: https://huggingface.co/datasets/SakanaAI/EDINET-Bench
--- a/yuholens-14b-Q3_K_M.gguf
+++ b/yuholens-14b-Q3_K_M.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50cd8dfe5a9de12cf01766ca5ac69be3b8cefab895e6fc4fc1c5020fc338817e
+size 7709891904
--- a/yuholens-14b-Q4_K_M.gguf
+++ b/yuholens-14b-Q4_K_M.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb99c3955ccb5b1d06008182c1605c9627212f5f8326292afc3b71c185990ff4
+size 9461361984
--- a/yuholens-14b-Q5_K_M.gguf
+++ b/yuholens-14b-Q5_K_M.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2fb3bd1d656d60b48169564a807aaab0148b995f2c688e68f36b38c3fa2ebf40
+size 10675252544
--- a/yuholens-14b-Q6_K.gguf
+++ b/yuholens-14b-Q6_K.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20fcadf3c501ef14ab98314801e2962a10c302f1ab4825b4d55bf2a406ca293a
+size 12310150464
--- a/yuholens-14b-Q8_0.gguf
+++ b/yuholens-14b-Q8_0.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92152f1d7f73c34112f37f7cfdea524bfc2b9d306ef21c052aa72f2bdaba98f4
+size 15061720384