You can deploy this model and start earning money today!

hfm8tr/
smollm2-135m-smoltalk-lora

Name: smollm2-135m-smoltalk-lora
Availability: InStock

HuggingFaceTB/SmolLM2-135M

adapter

Views

No views yet

SmolLM2-135M — Instruction Fine-Tune (LoRA)

A LoRA adapter for HuggingFaceTB/SmolLM2-135M
fine-tuned on 2,000 samples from HuggingFaceTB/smoltalk (everyday-conversations config).

This is a learning project — not a production model. It demonstrates that even 2k samples of
LoRA SFT teaches a base model the instruction-following format.

Training Details


Base model	SmolLM2-135M
Dataset	smoltalk/everyday-conversations (2k samples)
Epochs	1
LoRA r	8
LoRA alpha	16
Target modules	q_proj, v_proj
max_seq_length	1024
Compute	Google Colab T4 (~15 min)

How to Use

python

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

BASE_ID = "HuggingFaceTB/SmolLM2-135M"
ADAPTER_ID = "hfm8tr/smollm2-135m-smoltalk-lora"

tokenizer = AutoTokenizer.from_pretrained(ADAPTER_ID)
base = AutoModelForCausalLM.from_pretrained(BASE_ID, torch_dtype=torch.float32)
model = PeftModel.from_pretrained(base, ADAPTER_ID)
model.eval()

messages = [{"role": "user", "content": "What is gravity?"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt")

with torch.no_grad():
    out = model.generate(**inputs, max_new_tokens=100, do_sample=True, temperature=0.3,
                         pad_token_id=tokenizer.eos_token_id)
print(tokenizer.decode(out[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True))