Main Model •
Collection •
GitHub (MNN) •
Tech Report
English | 中文
MNN-Pocket-0.3B-SFT
The supervised fine-tuning (SFT) checkpoint of MNN-Pocket-0.3B — the intermediate stage between Base and the released MNN-Pocket-0.3B instruct model (which additionally applies the SFT Loop Engine and multi-teacher on-policy distillation). MNN-Pocket is a family of compact (~300M parameter) language models co-designed with the MNN inference stack for on-device deployment.
Find more details in the original model card: https://huggingface.co/taobao-mnn/MNN-Pocket-0.3B
- Released for reproducibility and research: ablate the post-training pipeline, or continue training from the SFT stage
- For general chat / assistant / on-device use, prefer the final instruct model MNN-Pocket-0.3B, which is strictly stronger (macro-average 50.00 vs the SFT stage on the 11-benchmark suite)
- Identical architecture and tokenizer to the instruct model; ChatML-style chat template included
Demo
iPhone 18 Pro Demo
Your browser does not support the video tag.Model Details
| Property | Value |
|---|---|
| Checkpoint | Supervised fine-tuned, intermediate stage |
| Parameters | ~300M |
| Architecture | Decoder-only transformer, 28 layers |
| Hidden size | 896 |
| Attention | 14 heads, GQA with 2 KV heads, head_dim 64 |
| FFN intermediate | 2560 (SiLU) |
| Context Length | 32K (40,960 position embeddings) |
| Vocabulary | 65,536 |
| Embeddings | tied input/output |
| Precision | bfloat16 |
| Languages | Chinese, English |
🏃 How to run
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "taobao-mnn/MNN-Pocket-0.3B-SFT"
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", dtype="bfloat16")
tokenizer = AutoTokenizer.from_pretrained(model_id)
text = tokenizer.apply_chat_template(
[{"role": "user", "content": "Hello!"}],
add_generation_prompt=True, tokenize=False)
Other Formats
- MNN-Pocket-0.3B — BF16 instruction-tuned, full model card (723 MB)
- MNN-Pocket-0.3B-Base — pre-trained base checkpoint
- MNN-Pocket-0.3B-GGUF / MNN-Pocket-0.3B-MNN — GGUF and INT4 MNN deployment builds
- MNN-Pocket-0.3B-Dec — non-autoregressive decision model (LoRA adapter + pointer head)
License
Apache License 2.0. See LICENSE.txt.