deepanshupillm commited on
Commit
31db0f5
·
verified ·
1 Parent(s): 58c0997

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +28 -10
README.md CHANGED
@@ -39,7 +39,25 @@ Alpie-Core is one of the world's first fine-tuned 4-bit reasoning models, provin
39
  - **License**: Apache 2.0
40
 
41
 
42
- ## 3. Model Features
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
43
 
44
  1. **Supports Streaming** – Real-time token-level responses
45
  2. **OpenAI-Compatible API** – Seamless integration with OpenAI client libraries
@@ -51,7 +69,7 @@ Alpie-Core is one of the world's first fine-tuned 4-bit reasoning models, provin
51
  8. **Customizable Safety & Moderation Filters** – Built-in guardrails for safer outputs
52
  9. **Supports Function Calling / Tool Use** – Enables structured outputs and external API integration
53
 
54
- ## 4. Key Highlights
55
 
56
  1. **Frontier Performance in 4-bit**: 81.28% MMLU, 92.75% GSM8K, 57.8% SWE-Bench Verified
57
 
@@ -65,7 +83,7 @@ Alpie-Core is one of the world's first fine-tuned 4-bit reasoning models, provin
65
 
66
  6) **Environmental Benefits**: Through quantization and efficiency-focused design, the model requires significantly fewer computational resources. This translates into lower energy consumption and reduced carbon footprint relative to full-precision deployments.
67
 
68
- ## 5. Benchmark Results
69
 
70
  | Benchmark | Alpie-Core (32B-4bit) | DeepSeek-V2 (236B) | Qwen2.5 72B | Llama 3.1 405B | Llama 3.1 70B | Gemma-3 27B-PT | Mistral-Small-24B-Base-2501 |
71
  |-----------|----------------------|-------------------|-------------|---------------|---------------|----------------|----------------------------|
@@ -115,7 +133,7 @@ Alpie-Core is one of the world's first fine-tuned 4-bit reasoning models, provin
115
  | AGIEval | **64.98%** | General Intelligence |
116
  | Winogrande | **79.53%** | Commonsense Reasoning |
117
 
118
- ## 6. Training Details
119
 
120
  - **Hardware**: 8× NVIDIA H100-80GB GPUs
121
  - **Training Duration**: 408 hours
@@ -127,11 +145,11 @@ Alpie-Core is one of the world's first fine-tuned 4-bit reasoning models, provin
127
  - **Dataset Domains**: Mathematics, coding, reasoning, science, general knowledge, competitive exams, Indian context + law, multilingual (Hindi and Hinglish)
128
  - **Synthetic Data Advantage**: +15-20% performance boost in STEM & coding domains
129
 
130
- ## 7. Environmental Impact
131
 
132
  **Carbon Footprint**: 298-835 kg CO₂e (training)
133
 
134
- ## 8. Use Cases
135
 
136
  Best for **STEM**, **complex mathematical reasoning**, **coding**, and **Indian context**
137
 
@@ -144,7 +162,7 @@ Best for **STEM**, **complex mathematical reasoning**, **coding**, and **Indian
144
  4)**Indian Context**: Provides culturally aware insights, competitive exam assistance (JEE, NEET, UPSC), and multilingual support in Hindi/Hinglish.
145
 
146
 
147
- ## 9. Safety and Limitations
148
 
149
  ### Enhanced Content Access
150
  Unlike the base DeepSeek model, Alpie-Core provides factual, balanced responses to geopolitically sensitive questions, offering global accessibility and factual accuracy on topics like Taiwan's status, Arunachal Pradesh sovereignty, and other sensitive geopolitical issues.
@@ -160,7 +178,7 @@ Unlike the base DeepSeek model, Alpie-Core provides factual, balanced responses
160
  - Model-assisted safety pipeline using RLHF
161
  - Comprehensive adversarial testing by domain experts
162
 
163
- ## 10. How to Use
164
 
165
  ### Non-Streaming Inference
166
  ```python
@@ -249,7 +267,7 @@ with torch.no_grad():
249
  - **vLLM**: High-throughput inference
250
  - **LMDeploy/Ollama/TensorRT-LLM**: Production deployments
251
 
252
- ## 11. Citation
253
 
254
  ```bibtex
255
  @misc{alpie2025core,
@@ -260,7 +278,7 @@ with torch.no_grad():
260
  }
261
  ```
262
 
263
- ## 12. License
264
 
265
  Apache 2.0 – Free for research and commercial use
266
 
 
39
  - **License**: Apache 2.0
40
 
41
 
42
+ ## 3. Approach
43
+
44
+ **Alpie-Core** has undergone extensive **supervised fine-tuning (SFT)** to strengthen reasoning, robustness, and safety. The training leveraged a diverse mixture of curated open-source datasets and proprietary synthetic data, optimized with high-quality LLM-generated responses. The fine-tuning process emphasized adherence to rigorous safety and usability standards, including:
45
+
46
+ **User Understanding and Clarity** – ensuring outputs are direct, interpretable, and pedagogically sound.
47
+
48
+ **Security and Ethical Guidelines** – filtering unsafe or harmful generations during and after training.
49
+
50
+ **Limitations, Disclaimers, and Knowledge Boundaries** – transparently communicating uncertainty and scope.
51
+
52
+ **Handling Complex and Sensitive Topics** – balancing informativeness with responsible guardrails.
53
+
54
+ **Safety and Respectful Engagement** – maintaining politeness, inclusivity, and cultural sensitivity.
55
+
56
+ **Confidentiality and Responsible Use** – preventing leakage of private training data, proprietary prompts, or internal reasoning traces.
57
+
58
+ This SFT approach enables Alpie-Core to deliver reliable, aligned, and context-aware responses while maintaining safety across a broad range of use cases.
59
+
60
+ ## 4. Model Features
61
 
62
  1. **Supports Streaming** – Real-time token-level responses
63
  2. **OpenAI-Compatible API** – Seamless integration with OpenAI client libraries
 
69
  8. **Customizable Safety & Moderation Filters** – Built-in guardrails for safer outputs
70
  9. **Supports Function Calling / Tool Use** – Enables structured outputs and external API integration
71
 
72
+ ## 5. Key Highlights
73
 
74
  1. **Frontier Performance in 4-bit**: 81.28% MMLU, 92.75% GSM8K, 57.8% SWE-Bench Verified
75
 
 
83
 
84
  6) **Environmental Benefits**: Through quantization and efficiency-focused design, the model requires significantly fewer computational resources. This translates into lower energy consumption and reduced carbon footprint relative to full-precision deployments.
85
 
86
+ ## 6. Benchmark Results
87
 
88
  | Benchmark | Alpie-Core (32B-4bit) | DeepSeek-V2 (236B) | Qwen2.5 72B | Llama 3.1 405B | Llama 3.1 70B | Gemma-3 27B-PT | Mistral-Small-24B-Base-2501 |
89
  |-----------|----------------------|-------------------|-------------|---------------|---------------|----------------|----------------------------|
 
133
  | AGIEval | **64.98%** | General Intelligence |
134
  | Winogrande | **79.53%** | Commonsense Reasoning |
135
 
136
+ ## 7. Training Details
137
 
138
  - **Hardware**: 8× NVIDIA H100-80GB GPUs
139
  - **Training Duration**: 408 hours
 
145
  - **Dataset Domains**: Mathematics, coding, reasoning, science, general knowledge, competitive exams, Indian context + law, multilingual (Hindi and Hinglish)
146
  - **Synthetic Data Advantage**: +15-20% performance boost in STEM & coding domains
147
 
148
+ ## 8. Environmental Impact
149
 
150
  **Carbon Footprint**: 298-835 kg CO₂e (training)
151
 
152
+ ## 9. Use Cases
153
 
154
  Best for **STEM**, **complex mathematical reasoning**, **coding**, and **Indian context**
155
 
 
162
  4)**Indian Context**: Provides culturally aware insights, competitive exam assistance (JEE, NEET, UPSC), and multilingual support in Hindi/Hinglish.
163
 
164
 
165
+ ## 10. Safety and Limitations
166
 
167
  ### Enhanced Content Access
168
  Unlike the base DeepSeek model, Alpie-Core provides factual, balanced responses to geopolitically sensitive questions, offering global accessibility and factual accuracy on topics like Taiwan's status, Arunachal Pradesh sovereignty, and other sensitive geopolitical issues.
 
178
  - Model-assisted safety pipeline using RLHF
179
  - Comprehensive adversarial testing by domain experts
180
 
181
+ ## 11. How to Use
182
 
183
  ### Non-Streaming Inference
184
  ```python
 
267
  - **vLLM**: High-throughput inference
268
  - **LMDeploy/Ollama/TensorRT-LLM**: Production deployments
269
 
270
+ ## 12. Citation
271
 
272
  ```bibtex
273
  @misc{alpie2025core,
 
278
  }
279
  ```
280
 
281
+ ## 13. License
282
 
283
  Apache 2.0 – Free for research and commercial use
284