Views
No views yet
openai/gpt-oss-20b base model using efficient fine-tuning techniques.pcm)pip install transformers torch accelerate1from transformers import AutoModelForCausalLM, AutoTokenizer
2import torch
3
4# Load model and tokenizer
5model = AutoModelForCausalLM.from_pretrained(
6 "Ephraimmm/pidgin_finetuned_model",
7 torch_dtype=torch.float16,
8 device_map="auto"
9)
10
11tokenizer = AutoTokenizer.from_pretrained("Ephraimmm/pidgin_finetuned_model")
12
13# Generate text
14prompt = "Wetin you dey do today?"
15inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
16
17outputs = model.generate(
18 **inputs,
19 max_new_tokens=100,
20 temperature=0.7,
21 top_p=0.9,
22 do_sample=True,
23 repetition_penalty=1.1
24)
25
26response = tokenizer.decode(outputs[0], skip_special_tokens=True)
27print(response)1from transformers import GenerationConfig
2
3# Create generation config for better control
4generation_config = GenerationConfig(
5 max_new_tokens=150,
6 temperature=0.8,
7 top_p=0.95,
8 top_k=50,
9 repetition_penalty=1.2,
10 do_sample=True,
11 pad_token_id=tokenizer.pad_token_id,
12 eos_token_id=tokenizer.eos_token_id,
13)
14
15# Generate with config
16outputs = model.generate(
17 **inputs,
18 generation_config=generation_config
19)
20
21text = tokenizer.decode(outputs[0], skip_special_tokens=True)
22print(text)1def generate_pidgin_response(prompt, model, tokenizer, max_length=100):
2 """Generate a Pidgin response to a prompt"""
3 inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
4
5 with torch.no_grad():
6 outputs = model.generate(
7 **inputs,
8 max_new_tokens=max_length,
9 temperature=0.7,
10 top_p=0.9,
11 do_sample=True,
12 repetition_penalty=1.1,
13 pad_token_id=tokenizer.pad_token_id,
14 )
15
16 return tokenizer.decode(outputs[0], skip_special_tokens=True)
17
18# Example conversation
19prompts = [
20 "How you dey?",
21 "Wetin you wan chop?",
22 "Make we go market?",
23 "I tire o!"
24]
25
26for prompt in prompts:
27 response = generate_pidgin_response(prompt, model, tokenizer)
28 print(f"Input: {prompt}")
29 print(f"Output: {response}\\n")"Wetin you dey do?""I dey here dey work small. You sef, how you dey?""How person go reach there?""You fit take bus from here, den you go come down for junction""One day, one man""One day, one man waka go market to buy something for him family. As e reach there, e see say..."1base_model: openai/gpt-oss-20b
2training_framework: unsloth
3fine_tuning_method: lora
4merged: true1# For limited VRAM, use 8-bit quantization
2from transformers import BitsAndBytesConfig
3
4quantization_config = BitsAndBytesConfig(
5 load_in_8bit=True,
6 llm_int8_threshold=6.0
7)
8
9model = AutoModelForCausalLM.from_pretrained(
10 "Ephraimmm/pidgin_finetuned_model",
11 quantization_config=quantization_config,
12 device_map="auto"
13)1@misc{pidgin_finetuned_model_2026,
2 author = {Ephraimmm},
3 title = {Pidgin English Fine-tuned Language Model},
4 year = {2026},
5 publisher = {HuggingFace},
6 journal = {HuggingFace Model Hub},
7 howpublished = {\\url{https://huggingface.co/Ephraimmm/pidgin_finetuned_model}}
8}