Qwen2.5-0.5B-Instruct (Customizable Copy)
This is a copy of Qwen/Qwen2.5-0.5B-Instruct for customization and fine-tuning.
π Model Details
- Base Model: Qwen/Qwen2.5-0.5B-Instruct
- Size: 0.5B parameters (~1GB)
- Type: Instruction-tuned language model
- License: Apache 2.0
π― Purpose
This repository contains a modifiable copy of Qwen 2.5 for:
- Fine-tuning on custom datasets
- Experimentation and testing
- RunPod serverless deployment
- Model modifications
π Usage
Direct Inference
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "marcosremar2/runpod_serverless_n2"
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
prompt = "What is artificial intelligence?"
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=512
)
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
RunPod Serverless Deployment
Environment Variables:
MODEL_NAME: marcosremar2/runpod_serverless_n2
HF_TOKEN: YOUR_TOKEN_HERE
MAX_MODEL_LEN: 4096
TRUST_REMOTE_CODE: true
GPU: RTX 4090 (24GB)
Min Workers: 0
Max Workers: 1
π§ Fine-tuning
To fine-tune this model:
from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, Trainer
model = AutoModelForCausalLM.from_pretrained("marcosremar2/runpod_serverless_n2")
tokenizer = AutoTokenizer.from_pretrained("marcosremar2/runpod_serverless_n2")
# Your fine-tuning code here
# ...
# Push back to your repo
model.push_to_hub("marcosremar2/runpod_serverless_n2")
tokenizer.push_to_hub("marcosremar2/runpod_serverless_n2")
π Performance
Metric | Value |
---|---|
Parameters | 0.5B |
Context Length | 32K tokens |
VRAM Required | ~1-2GB |
Inference Speed | 200-300 tokens/sec (RTX 4090) |
π Original Model
This is based on: Qwen/Qwen2.5-0.5B-Instruct
For more information about the Qwen2.5 series, visit the original repository.
π License
Apache 2.0 - Same as the original Qwen model.
π Credits
- Original Model: Qwen Team @ Alibaba Cloud
- Repository: Custom copy for modification and deployment
- Downloads last month
- 54