You can deploy this model and start earning money today!

thu-nmrc/
thaihao-100m-gpt-3b-sft6

Name: thaihao-100m-gpt-3b-sft6
Availability: InStock

text-generation-inference

endpoints_compatible

Views

No views yet

Thaihao 100M GPT 3B SFT6

This is a roughly 100M-parameter Chinese industrial business assistant prototype exported in Hugging Face Transformers GPT-2 causal language model format.

Intended Use

The model is intended for industrial business assistant scenarios such as:

field troubleshooting triage
acceptance checklist drafting
bid and tender technical response drafting
parameter verification reminders
safety and compliance reminders

It is not a general chat model and does not have live access to customer systems, weather, current work orders, quotes, contracts, or unpublished product data.

Model Details

Architecture: GPT-2 style causal LM
Parameters: about 95M tied-weight parameters
Context length: 512
Layers: 8
Attention heads: 8
Hidden size: 768
Vocabulary size: 50,000

Prompt Format

text

<|bos|>
<|user|> 用户：ATS 不切换怎么排查？
<|assistant|> 助手：

Python Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "thu-nmrc/thaihao-100m-gpt-3b-sft6"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = "<|bos|>\n<|user|> 用户：ATS 不切换怎么排查？\n<|assistant|> 助手："
inputs = tokenizer(prompt, return_tensors="pt")

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=160,
        do_sample=True,
        temperature=0.35,
        top_k=20,
        pad_token_id=tokenizer.pad_token_id,
    )

print(tokenizer.decode(output[0], skip_special_tokens=False))

Safety Notes

The model should not be used to disclose customer purchase quantities, quotations, contracts, contacts, work-order status, unpublished product parameters, military-sensitive details, or dangerous electrical operation steps. Use an authorized RAG or business system for current and permissioned data.