81 lines
4.0 KiB
Markdown
81 lines
4.0 KiB
Markdown
---
|
|
license: apache-2.0
|
|
datasets:
|
|
- openai/gsm8k
|
|
language:
|
|
- en
|
|
base_model:
|
|
- Qwen/Qwen2.5-0.5B-Instruct
|
|
pipeline_tag: text-generation
|
|
library_name: transformers
|
|
tags:
|
|
- math
|
|
- reasoning
|
|
- grpo
|
|
- trl
|
|
- code
|
|
---
|
|
# **Feynman-Grpo-Exp**
|
|
|
|
Feynman-Grpo-Exp is based on the Qwen 0.5B modality architecture, designed to enhance the reasoning capabilities of 0.5B-parameter models. It has been fine-tuned using the GRPO trainer on the OpenAI GSM8K dataset for reinforcement learning, improving its ability to handle complex reasoning tasks, multi-step problem-solving, and mathematical challenges. This model excels in chain-of-thought (CoT) reasoning and logical problem-solving, making it suitable for a variety of advanced tasks that require precise and structured outputs.
|
|
|
|
### **Key Improvements**
|
|
1. **Enhanced Knowledge and Expertise**: Strengthened mathematical reasoning, code generation, and problem-solving skills, particularly in scientific and technical domains.
|
|
2. **Fine-Tuned Instruction Following**: Optimized for generating structured outputs like JSON and handling long-form text (up to 8K+ tokens).
|
|
3. **Greater Adaptability**: Enhanced role-playing capabilities, allowing for better responses to diverse prompts.
|
|
4. **Long-Context Support**: Capable of processing up to **64K tokens** and generating up to **4K tokens** per output.
|
|
5. **Multilingual Proficiency**: Supports over **29 languages**, including Chinese, English, French, Spanish, Portuguese, German, and more.
|
|
|
|
### **Quickstart with Transformers**
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "prithivMLmods/Feynman-Grpo-Exp"
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto",
|
|
trust_remote_code=True
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
prompt = "Give me a short introduction to large language models."
|
|
messages = [
|
|
{"role": "system", "content": "You are an advanced AI assistant with expert-level reasoning and knowledge."},
|
|
{"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
|
|
)
|
|
generated_ids = [
|
|
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
|
]
|
|
|
|
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
print(response)
|
|
```
|
|
|
|
### **Intended Use**
|
|
- **Advanced Reasoning & Context Understanding**: Ideal for logical deduction, multi-step problem-solving, and complex knowledge-based tasks.
|
|
- **Mathematical & Scientific Problem-Solving**: Optimized for handling advanced calculations, theorem proving, and scientific queries.
|
|
- **Code Generation & Debugging**: Capable of generating and optimizing code across multiple programming languages.
|
|
- **Structured Data Analysis**: Processes structured data, including tables, JSON, and other formats, making it well-suited for data-centric tasks.
|
|
- **Multilingual Applications**: Proficient in over 29 languages, enabling a global scale for applications.
|
|
- **Extended Content Generation**: Supports detailed document writing, research reports, and instructional guides.
|
|
|
|
### **Limitations**
|
|
1. **Computational Requirements**: Despite being a **0.5B-parameter** model, it requires significant computational resources for efficient inference, especially when dealing with long-context processing.
|
|
2. **Language-Specific Variability**: Performance may vary across supported languages, with possible challenges for low-resource languages.
|
|
3. **Potential Error Accumulation**: Long-text generation can introduce inconsistencies or errors over extended outputs.
|
|
4. **Limited Real-World Awareness**: The model's knowledge is restricted to the training data, which may not reflect the most recent events or developments.
|
|
5. **Prompt Sensitivity**: Outputs depend heavily on the specificity and clarity of the input prompts. |