113 lines
3.9 KiB
Markdown
113 lines
3.9 KiB
Markdown
---
|
||
license: apache-2.0
|
||
datasets:
|
||
- GeneralReasoning/GeneralThought-430K
|
||
base_model:
|
||
- prithivMLmods/Qwen3-4B-ft-bf16
|
||
language:
|
||
- en
|
||
pipeline_tag: text-generation
|
||
library_name: transformers
|
||
tags:
|
||
- moe
|
||
- text-generation-inference
|
||
- code
|
||
- math
|
||
---
|
||
|
||
|
||
|
||
# Cetus-Qwen3\_4B-GeneralThought
|
||
|
||
> Cetus-Qwen3\_4B-GeneralThought is a fine-tuned variant of the Qwen3-4B architecture, trained on the GeneralThought-430K dataset to enhance broad-spectrum reasoning, logical coherence, and structured multi-domain problem solving. This model is optimized for general-purpose tasks including instruction following, technical question answering, and reasoning-based generation across diverse knowledge fields.
|
||
|
||
|
||
> [!note]
|
||
[ GGUF ] : https://huggingface.co/prithivMLmods/Cetus-Qwen3_4B-GeneralThought-Q4_K_M-GGUF
|
||
|
||
## Key Features
|
||
|
||
1. Broad Reasoning with GeneralThought-430K
|
||
Trained on a carefully curated 430,000-sample dataset—GeneralThought-430K—spanning:
|
||
|
||
* Mathematical and logical reasoning
|
||
* Scientific and factual QA
|
||
* Multistep instructions and problem decomposition
|
||
* Abstract and applied reasoning tasks
|
||
|
||
2. Multi-Domain Task Versatility
|
||
Equipped to handle use cases across STEM, humanities, code reasoning, and general knowledge workflows with consistency and structure.
|
||
|
||
3. Structured Output Control
|
||
Outputs well-formatted answers in Markdown, LaTeX, JSON, and tabular formats, suitable for documentation, education, and technical reporting.
|
||
|
||
4. Instruction-Following with Multi-Step Fidelity
|
||
Capable of following detailed prompts involving layered reasoning or procedural guidance with high step-to-step coherence.
|
||
|
||
5. Multilingual and Cross-Cultural Understanding
|
||
Supports over 20 languages for global comprehension tasks and technical translation in education and public sector applications.
|
||
|
||
6. Efficient Qwen3-4B Base
|
||
Offers an optimal balance between intelligence and computational efficiency—ideal for deployment on consumer-grade GPUs and scalable services.
|
||
|
||
## Quickstart with Transformers
|
||
|
||
```python
|
||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
||
model_name = "prithivMLmods/Cetus-Qwen3_4B-GeneralThought"
|
||
|
||
model = AutoModelForCausalLM.from_pretrained(
|
||
model_name,
|
||
torch_dtype="auto",
|
||
device_map="auto"
|
||
)
|
||
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||
|
||
prompt = "Explain the concept of entropy in thermodynamics in simple terms."
|
||
|
||
messages = [
|
||
{"role": "system", "content": "You are a general-purpose reasoning assistant trained on GeneralThought-430K."},
|
||
{"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
|
||
|
||
* General reasoning and educational Q\&A
|
||
* Technical concept explanation and summarization
|
||
* Structured content generation in Markdown, LaTeX, and JSON
|
||
* Code and logic support in instruction-rich workflows
|
||
* Multi-language academic and public knowledge tools
|
||
|
||
## Limitations
|
||
|
||
* Not optimized for purely creative or fictional content
|
||
* Smaller context window compared to frontier models
|
||
* May be sensitive to ambiguous or poorly structured prompts
|
||
* Can occasionally hallucinate in niche or adversarial scenarios
|
||
|
||
## References
|
||
|
||
1. Qwen2.5 Technical Report – [https://arxiv.org/pdf/2412.15115](https://arxiv.org/pdf/2412.15115)
|
||
2. YaRN: Context Window Extension – [https://arxiv.org/pdf/2309.00071](https://arxiv.org/pdf/2309.00071)
|
||
3. GeneralThought-430K Dataset – (internal/prepublication dataset source, if applicable) |