---
base_model: Qwen/Qwen3-4B-Base
library_name: peft
license: other
language:
  - en
tags:
  - lora
  - peft
  - adapter
  - adaption
  - autoscientist
---

# adaption_fim_event_classification

## Model Training

A LORA adapter for `Qwen/Qwen3-4B-Base`. This model was trained with SFT using [Adaption](https://adaptionlabs.ai)'s AutoScientist on the fim_event_classification dataset.


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

### AutoScientist Config

```json
{
  "job_id": "15245b3b-55a2-4540-9b8a-9ff40a459d60",
  "training_experiment_id": "9642fe2d-3d48-4612-86cc-d9239c844595",
  "original_model_name": "Qwen/Qwen3-4B-Base",
  "trained_model_name": "adaption_fim_event_classification",
  "training_method": "sft",
  "training_type": "lora",
  "data_format": "chat",
  "hyperparams": {
    "lora": "true",
    "lora_r": 32,
    "n_evals": 5,
    "n_epochs": 5,
    "batch_size": "max",
    "lora_alpha": 64,
    "lora_dropout": 0.05,
    "min_lr_ratio": 0.2,
    "warmup_ratio": 0.05,
    "weight_decay": 0.01,
    "learning_rate": 0.0001,
    "max_grad_norm": 1,
    "base_model_size": "4B",
    "train_on_inputs": "false",
    "training_method": "sft",
    "lr_scheduler_type": "cosine",
    "scheduler_num_cycles": 0.5,
    "lora_trainable_modules": "k_proj,up_proj,o_proj,q_proj,down_proj,v_proj,gate_proj"
  }
}
```

## Training Data

The model was trained on 27,254 rows of adapted data with the following domain distribution: technology (71%), science (5%), academic-education (4%), governance (2%), math (2%), history (2%), corporate-business (2%), code (1%), writing-editing-communication (1%), entertainment (1%), cooking (1%), culture (1%), geography (1%), sports (1%), language (1%), travel (1%), fitness-sports (1%), music (0%), medical (0%), how-to (0%), product-advice (0%), animal-nature (0%), religion (0%), transportation (0%), legal (0%), personal-finance (0%), data-analysis-visualization (0%), personal-growth (0%), career-workplace (0%), games (0%), agriculture (0%), parenting-family (0%), art (0%), news (0%), architecture-design (0%), dating (0%), fashion-beauty (0%), marketing (0%), social (0%), hr (0%), literature (0%), roleplay (0%), market-analysis (0%), other (0%), outdoor-recreation (0%), media (0%), beauty (0%), philosophy (0%), survival-skills (0%), humanitarian-aid (0%), survival-outdoors (0%), family-relationships (0%), academic-游戏副本 (0%), survival-preparedness (0%).

## 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)

| Domain | Win rate vs. base model |
| --- | --- |
| technology | 71% |

## How to use

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

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

BASE = "Qwen/Qwen3-4B-Base"
ADAPTER = "dowdryan/adaption_fim_event_classification"

base = AutoModelForCausalLM.from_pretrained(BASE, dtype="auto", device_map="auto")
# Merge the LoRA weights into the base for faster inference
model = PeftModel.from_pretrained(base, ADAPTER).merge_and_unload()
model.eval()
tokenizer = AutoTokenizer.from_pretrained(BASE)

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

messages = [{"role": "user", "content": "Hello!"}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])
```