update metadata
Browse files
README.md
CHANGED
@@ -1,3 +1,109 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
base_model:
|
6 |
+
- Qwen/Qwen3-0.6B
|
7 |
+
pipeline_tag: text-generation
|
8 |
+
library_name: transformers
|
9 |
+
tags:
|
10 |
+
- text-generation-inference
|
11 |
+
- DAG
|
12 |
+
- gspo
|
13 |
+
- trl
|
14 |
+
- math
|
15 |
+
- code
|
16 |
+
---
|
17 |
+
|
18 |
+
# **Telescopium-Acyclic-Qwen3-0.6B**
|
19 |
+
|
20 |
+
> **Telescopium-Acyclic-Qwen3-0.6B** is a high-efficiency, multi-domain model fine-tuned on **Qwen-0.6B** using the **rStar-Coder** dataset enhanced with **code expert clusters** and an extended **open code reasoning dataset**, plus **deepseek-r1 math reasoning traces**. It leverages **Directed Acyclic Graph (DAG) multistep reasoning** for precise symbolic problem solving in mathematics, code, and science—making it ideal for developers, educators, and researchers working with structured reasoning pipelines under constrained compute.
|
21 |
+
|
22 |
+
> \[!note]
|
23 |
+
> GGUF: [https://huggingface.co/prithivMLmods/Telescopium-Acyclic-Qwen3-0.6B-GGUF](https://huggingface.co/prithivMLmods/Telescopium-Acyclic-Qwen3-0.6B-GGUF)
|
24 |
+
|
25 |
+
---
|
26 |
+
|
27 |
+
## **Key Features**
|
28 |
+
|
29 |
+
1. **DAG-Based Multistep Reasoning for Math**
|
30 |
+
Implements **Directed Acyclic Graph (DAG) reasoning methodology** to break down complex mathematical problems into dependency-ordered steps, inspired by **deepseek-r1 reasoning traces**.
|
31 |
+
|
32 |
+
2. **Unified Reasoning Across Code, Math & Science**
|
33 |
+
Fine-tuned on **expert clusters** spanning programming, mathematics, and scientific logic, alongside an **open code reasoning dataset**, enabling cross-domain symbolic precision.
|
34 |
+
|
35 |
+
3. **Advanced Code Reasoning & Generation**
|
36 |
+
Supports multi-language coding with explanations, optimization hints, and error detection—ideal for full-stack prototyping, algorithm synthesis, and debugging workflows.
|
37 |
+
|
38 |
+
4. **Scientific Problem Solving**
|
39 |
+
Performs analytical reasoning in physics, biology, and chemistry—explaining concepts, solving equations, and handling symbolic derivations step-by-step.
|
40 |
+
|
41 |
+
5. **Hybrid Symbolic-AI Thinking**
|
42 |
+
Combines **DAG logic decomposition**, chain-of-thought reasoning, and open-ended inference, delivering robust performance on STEM tasks and complex prompt decomposition.
|
43 |
+
|
44 |
+
6. **Structured Output Mastery**
|
45 |
+
Seamlessly generates output in **LaTeX**, **Markdown**, **JSON**, **CSV**, and **YAML**, suited for research reports, technical documentation, and data formats.
|
46 |
+
|
47 |
+
7. **Optimized Lightweight Footprint for Versatile Deployment**
|
48 |
+
Strikes a balance between performance and efficiency, making it deployable on **mid-range GPUs**, **offline clusters**, and advanced **edge AI systems**.
|
49 |
+
|
50 |
+
---
|
51 |
+
|
52 |
+
## **Quickstart with Transformers**
|
53 |
+
|
54 |
+
```python
|
55 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
56 |
+
|
57 |
+
model_name = "prithivMLmods/Telescopium-Acyclic-Qwen3-0.6B"
|
58 |
+
|
59 |
+
model = AutoModelForCausalLM.from_pretrained(
|
60 |
+
model_name,
|
61 |
+
torch_dtype="auto",
|
62 |
+
device_map="auto"
|
63 |
+
)
|
64 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
65 |
+
|
66 |
+
prompt = "Solve the equation: 3x^2 + 5x - 2 = 0 using DAG-based step decomposition."
|
67 |
+
|
68 |
+
messages = [
|
69 |
+
{"role": "system", "content": "You are a STEM reasoning tutor using DAG multistep methodology for problem solving."},
|
70 |
+
{"role": "user", "content": prompt}
|
71 |
+
]
|
72 |
+
|
73 |
+
text = tokenizer.apply_chat_template(
|
74 |
+
messages,
|
75 |
+
tokenize=False,
|
76 |
+
add_generation_prompt=True
|
77 |
+
)
|
78 |
+
|
79 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
80 |
+
|
81 |
+
generated_ids = model.generate(
|
82 |
+
**model_inputs,
|
83 |
+
max_new_tokens=512
|
84 |
+
)
|
85 |
+
generated_ids = [
|
86 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
87 |
+
]
|
88 |
+
|
89 |
+
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
90 |
+
print(response)
|
91 |
+
```
|
92 |
+
|
93 |
+
---
|
94 |
+
|
95 |
+
## **Intended Use**
|
96 |
+
|
97 |
+
* Mathematical tutoring with **DAG-based decomposition**
|
98 |
+
* Scientific and computational logic education
|
99 |
+
* Advanced coding assistant for algorithm design, code reviews, and documentation
|
100 |
+
* Structured technical data generation across formats and fields
|
101 |
+
* STEM-focused chatbot or API for research and education tools
|
102 |
+
* Mid-resource deployment requiring high symbolic fidelity
|
103 |
+
|
104 |
+
## **Limitations**
|
105 |
+
|
106 |
+
* Not tuned for general-purpose or long-form creative writing
|
107 |
+
* Context limitations may hinder multi-document or full codebase analysis
|
108 |
+
* Specialized in technical and symbolic tasks—general chat may underperform
|
109 |
+
* Prioritizes structured reasoning over emotional or casual tone generation
|