# twinkle-ai/gemma-3-4B-T1-it

Model from Hugging Face for text generation: twinkle-ai/gemma-3-4B-T1-it · transformers

## Facts

- Type: LLM
- Task: text-generation
- Framework: transformers
- License: gemma — Commercial use is permitted with obligations: Gemma terms: commercial use allowed subject to Google's prohibited-use policy.
- Source: https://huggingface.co/twinkle-ai/gemma-3-4B-T1-it
- Hosting: hosted (9 files)
- Files: 9 hosted of 14 listed
- File counts: "listed" is every file in the source repository; only hosted files download from Hugging Bay, and the download plan lists the files it selects for one download, so its count can differ.
- Hosted here: 9 files, 9.14 GB
- Full repo upstream: 9.14 GB
- Recommended file: 9.1 GB (2 files)
- Parameters: 4B
- Download: https://huggingbay.xyz/artifact/hf-model-twinkle-ai-gemma-3-4b-t1-it/files (download plan: https://huggingbay.xyz/api/artifacts/hf-model-twinkle-ai-gemma-3-4b-t1-it/download-plan)
- Per-file SHA-256: available for 9 of 9 hosted files via https://huggingbay.xyz/api/artifacts/hf-model-twinkle-ai-gemma-3-4b-t1-it/files?limit=100
- Run-ready: no (Hugging Bay has not certified a run path for this model; hosted bytes are downloadable and hash-checkable only)
- Peer-assisted fallbacks: 0 verified, 0 pending review
- Downloads: 50502
- Stars: 41
- Card completeness: 96%

## Description

External Hugging Face catalog record for twinkle-ai/gemma-3-4B-T1-it. Task: text generation. Library: transformers. Languages: en, zh. Datasets referenced by card metadata: lianghsun/tw-reasoning-instruct, lianghsun/tw-contract-review-chat, minyichen/tw-instruct-R1-200k, minyichen/tw_mm_R1, minyichen/LongPaper_multitask_zh_tw_R1, nvidia/Nemotron-Instruction-Following-Chat-v1. Includes 3 model-card evaluation signals. Popularity signal at import: 50,502 downloads and 41 likes.

## Model Card Facts

- Declared license: gemma
- Pipeline: text-generation
- Library: transformers
- Task categories: text-generation
- Languages: en, zh
- Referenced datasets: lianghsun/tw-reasoning-instruct, lianghsun/tw-contract-review-chat, minyichen/tw-instruct-R1-200k, minyichen/tw_mm_R1, minyichen/LongPaper_multitask_zh_tw_R1, nvidia/Nemotron-Instruction-Following-Chat-v1
- Metrics: accuracy
- Model-index entries: gemma-3-4B-T1-it
- Evaluation signals: 3 model-card eval results
- File hints: 14 upstream file samples

## Evaluation Signals

- tmmlu+: single choice = 47.44 (Single Choice Question)
- mmlu: single choice = 59.13 (Single Choice Question)
- tw-legal-benchmark-v1: single choice = 44.18 (Single Choice Question)

## Referenced Data

- Dataset: lianghsun/tw-reasoning-instruct
- Dataset: lianghsun/tw-contract-review-chat
- Dataset: minyichen/tw-instruct-R1-200k
- Dataset: minyichen/tw_mm_R1
- Dataset: minyichen/LongPaper_multitask_zh_tw_R1
- Dataset: nvidia/Nemotron-Instruction-Following-Chat-v1

## File Hints

- .gitattributes (1.6 KB)
- .gitignore (39 B)
- README.md (18 KB)
- README_EN.md (18 KB)
- added_tokens.json (35 B)
- config.json (1.7 KB)
- generation_config.json (208 B)
- model-00001-of-00002.safetensors (4.6 GB) LFS
- model-00002-of-00002.safetensors (3.9 GB) LFS
- model.safetensors.index.json (36 KB)
- special_tokens_map.json (670 B)
- tokenizer.json (32 MB) LFS

## Quality Checklist

- Summary: passed
- Description: passed
- Model card metadata: passed
- License: review
- Provenance: passed
- File inventory: passed
- Hosted bytes: passed
- Scan result: passed
- Manifest signature: passed

## Safety Notes

- External catalog metadata imported from Hugging Face public API on 2026-09-21T06:29:44.680Z.
- Upstream license: gemma. Mirror admission policy: permissive (owner policy; no license rights inferred).
- Model-index metadata found for gemma-3-4B-T1-it.
- Model card references dataset context: lianghsun/tw-reasoning-instruct, lianghsun/tw-contract-review-chat, minyichen/tw-instruct-R1-200k, minyichen/tw_mm_R1.
- Not a run kit: hosted, hash-checked bytes only; run readiness is not certified for this model yet.

## Actions

- Web: https://huggingbay.xyz/artifact/hf-model-twinkle-ai-gemma-3-4b-t1-it
- Files API: https://huggingbay.xyz/api/artifacts/hf-model-twinkle-ai-gemma-3-4b-t1-it/files?limit=100
- Metadata: https://huggingbay.xyz/api/artifacts/hf-model-twinkle-ai-gemma-3-4b-t1-it/metadata.json
- Distribution: https://huggingbay.xyz/api/artifacts/hf-model-twinkle-ai-gemma-3-4b-t1-it/distribution
- Manifest: https://huggingbay.xyz/api/manifests/hf-model-twinkle-ai-gemma-3-4b-t1-it
- Download plan (per-file SHA-256 + signed download URLs): https://huggingbay.xyz/api/artifacts/hf-model-twinkle-ai-gemma-3-4b-t1-it/download-plan
- Hosted download rows: https://huggingbay.xyz/api/downloadable-files?artifactId=hf-model-twinkle-ai-gemma-3-4b-t1-it&limit=100
- Request mirror: not applicable.
- Submit peer fallback for review: POST https://huggingbay.xyz/api/artifacts/hf-model-twinkle-ai-gemma-3-4b-t1-it/decentralized-fallbacks