You can deploy this model and start earning money today!

Dapinsky/
PIT-4B-FT-202212-math-grpo-lora

Name: PIT-4B-FT-202212-math-grpo-lora
Availability: InStock

Diamegs/PIT-4B-FT-202212

adapter

Views

No views yet

Dapinsky/PIT-4B-FT-202212-math-grpo-lora

This repo contains a PEFT LoRA adapter trained from Diamegs/PIT-4B-FT-202212. It intentionally does not include merged base-model weights.

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = "Diamegs/PIT-4B-FT-202212"
adapter_id = "Dapinsky/PIT-4B-FT-202212-math-grpo-lora"
tokenizer = AutoTokenizer.from_pretrained(adapter_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(base_model, trust_remote_code=True)
model = PeftModel.from_pretrained(model, adapter_id)

Training Summary

Base model: Diamegs/PIT-4B-FT-202212
Fine-tuning method: LoRA post-training, adapter-only upload.
Tokenizer files copied from: base model

LoRA Configuration

json

{
  "peft_type": "LORA",
  "task_type": "CAUSAL_LM",
  "r": 64,
  "lora_alpha": 128,
  "lora_dropout": 0.05,
  "target_modules": [
    "c_q",
    "c_k",
    "c_proj",
    "c_v",
    "c_fc"
  ],
  "base_model_name_or_path": "Diamegs/PIT-4B-FT-202212"
}