thaihao-100m-gpt-sft7-default-reply

Thaihao 100M GPT SFT7 Default Reply

This is a local from-scratch Thaihao industrial business assistant checkpoint, exported to the Hugging Face GPT-2 causal LM format.

Model

• Parameters: 102,508,800 tied-weight parameters
• Source run: D:\thaihao\pilot_100m\runs\thaihao_100m_full_sft7_default_reply_20260519_110300
• Source checkpoint: D:\thaihao\pilot_100m\runs\thaihao_100m_full_sft7_default_reply_20260519_110300\checkpoints\final.pt
• Architecture: GPT-2 style causal LM
• Context length: 512
• Layers: 9
• Heads: 8
• Hidden size: 768
• Vocab size: 50000

Default Identity Reply

我是泰豪工业业务助手。可以帮你整理现场排查思路、验收清单、招投标技术响应、参数核对和安全合规提醒。我的能力偏工业业务，不是通用闲聊大模型；遇到实时数据或授权资料不足时，我会说明不能确定。

Prompt Format

<|bos|>
<|user|> 用户：...
<|assistant|> 助手：

Python Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "YOUR_ACCOUNT/thaihao-100m-gpt-sft7-default-reply"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = '<|bos|>\n<|user|> 用户：ATS 不切换怎么办？\n<|assistant|> 助手：'
inputs = tokenizer(prompt, return_tensors="pt")
with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=120,
        do_sample=True,
        temperature=0.45,
        top_k=30,
        pad_token_id=tokenizer.pad_token_id,
    )
print(tokenizer.decode(output[0], skip_special_tokens=False))

Important Limitations

This is a roughly 100M-parameter, limited-data business assistant prototype, not a 120B-token commercial general-purpose LLM. For customer names, quotations, contracts, work-order status, unpublished models, military-sensitive information or dangerous operations, connect an authorized RAG/business system and enforce permission checks.