---
base_model: google/gemma-3-4b-it
library_name: peft
license: other
tags:
  - lora
  - peft
  - adapter
  - adaption
---

# adaption_codeintel_python_reasoning_v

## Model Training

A LORA adapter for `google/gemma-3-4b-it`. This model was trained with SFT using [Adaption](https://adaptionlabs.ai)'s AutoScientist on the CodeIntel-Python-Reasoning-v1 dataset.


![Training metrics](training-metrics.png)

### AutoScientist Config

```json
{
  "job_id": "23277988-2422-4c04-8552-1bd3d6c06ff3",
  "training_experiment_id": "486c999b-f867-47c4-b288-e9e5624c0aad",
  "original_model_name": "google/gemma-3-4b-it",
  "trained_model_name": "adaption_codeintel_python_reasoning_v",
  "training_method": "sft",
  "training_type": "lora",
  "data_format": "chat",
  "hyperparams": {
    "lora": "true",
    "lora_r": 32,
    "n_evals": 5,
    "n_epochs": 1,
    "batch_size": "max",
    "lora_alpha": 64,
    "lora_dropout": 0,
    "min_lr_ratio": 0.1,
    "warmup_ratio": 0.1,
    "weight_decay": 0,
    "learning_rate": 0.00001,
    "max_grad_norm": 2,
    "base_model_size": "4B",
    "train_on_inputs": "false",
    "training_method": "sft",
    "lr_scheduler_type": "cosine",
    "scheduler_num_cycles": 0.5,
    "lora_trainable_modules": "all-linear"
  }
}
```

## Training Data

The model was trained on 27,678 rows of adapted data with the following domain distribution: code (97%), math (3%).

## Model Evaluation

The model was evaluated on an in-distribution held-out test set as well as a broader domain-specific test set to measure generalization.


![Win rates](win-rates.png)


## How to use

```bash
pip install torch transformers peft
```

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

BASE = "google/gemma-3-4b-it"
ADAPTER = "<this-repo-id>"

device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float32 if device == "cpu" else torch.bfloat16

base = AutoModelForCausalLM.from_pretrained(BASE, dtype=dtype).to(device)
model = PeftModel.from_pretrained(base, ADAPTER)
# Optional: merge the LoRA weights into the base for faster inference
model = model.merge_and_unload()
model.eval()

tokenizer = AutoTokenizer.from_pretrained(BASE)
messages = [{"role": "user", "content": "Hello!"}]
text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(device)

with torch.inference_mode():
    out = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
```