File size: 3,455 Bytes
7c9885b 52c8a7b 3ac5143 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 |
---
license: apache-2.0
datasets:
- microsoft/orca-math-word-problems-200k
- ise-uiuc/Magicoder-Evol-Instruct-110K
- Vezora/Tested-22k-Python-Alpaca
---
# Datacard for Custom Trained Model
- Base Model : [Kukedlc/NeuralExperiment-7b-dare-ties](https://huggingface.co/Kukedlc/NeuralExperiment-7b-dare-ties)
## Model Description
This model is an experimental AI trained on three distinct datasets focusing on logical reasoning, mathematics, and programming. The training process involved fine-tuning from the last layer (31) backward with a gradually decreasing learning rate. The primary goal is to address and rectify the common 'INSTINST' bug observed in leaderboard models through targeted training on the latest layers.
## Datasets Used for Training
- `microsoft/orca-math-word-problems-200k`: A large-scale dataset of mathematical word problems aimed at enhancing the model's numerical reasoning and problem-solving capabilities.
- `ise-uiuc/Magicoder-Evol-Instruct-110K`: A dataset designed to improve code generation and understanding, contributing to the model's programming language proficiency.
- `sahil2801/CodeAlpaca-20k`: A dataset focused on programming challenges to further refine the model's coding and logical reasoning skills.
Each dataset contributed 20,000 data points to the training process, ensuring a balanced representation of logic, mathematics, and programming tasks.
## Training Environment
- The model was trained on Kaggle's free GPU environment, allowing for cost-effective fine-tuning and experimentation.
- Users interested in replicating or extending this training can find the Kaggle notebook in my profile or request it directly for collaborative purposes.
## Preliminary Results
- The model shows promising results in solving logical puzzles and mathematical problems, especially those with misleading or non-obvious solutions that it initially struggled with.
- Ongoing experiments aim to quantify the impact of targeted training on the model's reasoning capabilities across different domains.
## Invitation for Collaboration
- Feedback, suggestions, and collaborative efforts are highly encouraged to further refine and evaluate the model.
- If interested in contributing or experimenting with this model, please feel free to reach out or access the code directly from my Kaggle profile.
## Contact Information
- For any inquiries, suggestions, or collaboration proposals, please contact me!
```python
!pip install -qU transformers accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "Kukedlc/NeuralExperiment-7b-MagicCoder-v7"
messages = [{"role": "user", "content": "What is a large language model?"}]
tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
"text-generation",
model=model,
torch_dtype=torch.float16,
device_map="auto",
)
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
```
|