RehanKingggg commited on
Commit
eb48b92
·
verified ·
1 Parent(s): 323c97e

Delete model_card.md

Browse files
Files changed (1) hide show
  1. model_card.md +0 -275
model_card.md DELETED
@@ -1,275 +0,0 @@
1
- ---
2
- license: other
3
- language:
4
- - en
5
- - zh
6
- library_name: transformers
7
- pipeline_tag: text-generation
8
- tags:
9
- - reasoning
10
- - mathematics
11
- - programming
12
- - creative-writing
13
- - chain-of-thought
14
- ---
15
-
16
- # Brello Thinking
17
-
18
- ## Model Description
19
-
20
- **Brello Thinking** is an advanced language model created by **Epic Systems** and engineered by **Rehan Temkar**. Built on the robust Tencent Hunyuan base model, Brello Thinking specializes in deep reasoning, mathematical problem-solving, coding, and creative thinking with enhanced chain-of-thought capabilities.
21
-
22
- ### Key Features
23
-
24
- - **🧠 Advanced Reasoning**: Enhanced chain-of-thought capabilities with both fast and slow thinking modes
25
- - **🔢 Mathematical Excellence**: Superior performance in mathematical problem-solving and computation
26
- - **💻 Programming Prowess**: Strong coding capabilities across multiple programming languages
27
- - **📚 Long Context Understanding**: Supports extended conversations and document analysis
28
- - **🎯 Creative Problem Solving**: Innovative approaches to complex problems
29
- - **🌐 Multi-language Support**: Fluent in multiple languages with cultural understanding
30
-
31
- ## Model Architecture
32
-
33
- - **Base Model**: Tencent Hunyuan
34
- - **Parameters**: 1.8B (optimized for efficiency)
35
- - **Context Window**: 256K tokens
36
- - **Architecture**: EpicBrelloV1ForCausalLM
37
- - **Specialization**: Reasoning, Mathematics, Programming, Creative Thinking
38
-
39
- ## Model Summary
40
-
41
- | Specification | Value |
42
- |---------------|-------|
43
- | **Architecture** | EpicBrelloV1ForCausalLM |
44
- | **Total Parameters** | 1.8B |
45
- | **Context Window** | 256K tokens |
46
- | **Hidden Size** | 2048 |
47
- | **Number of Layers** | 32 |
48
- | **Number of Attention Heads** | 16 |
49
- | **Number of Key-Value Heads** | 4 |
50
- | **Intermediate Size** | 6144 |
51
- | **Vocabulary Size** | 120,818 |
52
- | **Activation Function** | SiLU |
53
- | **Normalization** | RMSNorm |
54
- | **Position Embedding** | RoPE (Rotary Position Embedding) |
55
- | **Base Model** | Tencent Hunyuan |
56
- | **Creator** | Epic Systems |
57
- | **Engineer** | Rehan Temkar |
58
- | **License** | Proprietary - Epic Systems |
59
-
60
- ## Performance Benchmarks
61
-
62
- ### Mathematical Reasoning
63
- - **GSM8K**: 87.49% (Mathematical problem-solving)
64
- - **MATH**: 72.25% (Advanced mathematics)
65
- - **AIME 2024**: 78.3% (American Invitational Mathematics Examination)
66
- - **AIME 2025**: 66.5% (American Invitational Mathematics Examination)
67
-
68
- ### Programming & Coding
69
- - **MBPP**: 76.46% (Python programming)
70
- - **MultiPL-E**: 59.87% (Multi-language programming)
71
- - **LiveCodeBench**: 49.4% (Live coding evaluation)
72
- - **FullStackBench**: 54.6% (Full-stack development)
73
-
74
- ### Reasoning & Problem Solving
75
- - **BBH**: 75.17% (Big-Bench Hard tasks)
76
- - **DROP**: 78.2% (Reading comprehension)
77
- - **ZebraLogic**: 83.5% (Logical reasoning)
78
- - **CRUX-O**: 56.5% (Complex reasoning)
79
-
80
- ### Agent & Tool Use
81
- - **BFCL v3**: 67.9% (Benchmark for Function Calling)
82
- - **τ-Bench**: 30.3% (Tool use evaluation)
83
- - **ComplexFuncBench**: 26.3% (Complex function calling)
84
- - **C3-Bench**: 64.3% (Code comprehension)
85
-
86
- ### Long Context Understanding
87
- - **PenguinScrolls**: 83.1% (Long document processing)
88
- - **LongBench-v2**: 44.1% (Long context evaluation)
89
- - **FRAMES**: 79.2% (Frame-based reasoning)
90
-
91
- ## Usage
92
-
93
- ### Basic Usage
94
-
95
- ```python
96
- from transformers import AutoModelForCausalLM, AutoTokenizer
97
-
98
- # Load Brello Thinking
99
- model_name = "BrelloES/brello-thinking"
100
- tokenizer = AutoTokenizer.from_pretrained(model_name)
101
- model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")
102
-
103
- # Example conversation
104
- messages = [
105
- {"role": "user", "content": "What is 2+2?"}
106
- ]
107
-
108
- tokenized_chat = tokenizer.apply_chat_template(
109
- messages,
110
- tokenize=True,
111
- add_generation_prompt=True,
112
- return_tensors="pt",
113
- enable_thinking=True
114
- )
115
-
116
- outputs = model.generate(
117
- tokenized_chat.to(model.device),
118
- max_new_tokens=2048,
119
- do_sample=True,
120
- top_k=20,
121
- top_p=0.8,
122
- repetition_penalty=1.05,
123
- temperature=0.7
124
- )
125
-
126
- response = tokenizer.decode(outputs[0])
127
- print(response)
128
- ```
129
-
130
- ### Thinking Mode
131
-
132
- Brello Thinking supports enhanced reasoning with thinking mode:
133
-
134
- ```python
135
- # Enable thinking mode (default)
136
- tokenized_chat = tokenizer.apply_chat_template(
137
- messages,
138
- enable_thinking=True # Shows reasoning process
139
- )
140
-
141
- # Disable thinking mode
142
- tokenized_chat = tokenizer.apply_chat_template(
143
- messages,
144
- enable_thinking=False # Direct answers
145
- )
146
- ```
147
-
148
- ### Advanced Usage with System Prompts
149
-
150
- ```python
151
- # Custom system prompt for Brello Thinking
152
- system_prompt = """You are Brello Thinking, an advanced AI assistant created by Epic Systems and engineered by Rehan Temkar. You are built on the Tencent Hunyuan base model and specialize in deep reasoning, mathematical problem-solving, coding, and creative thinking. You have enhanced chain-of-thought capabilities and can show your thinking process."""
153
-
154
- messages = [
155
- {"role": "system", "content": system_prompt},
156
- {"role": "user", "content": "Solve this math problem: 2x + 5 = 13"}
157
- ]
158
-
159
- tokenized_chat = tokenizer.apply_chat_template(
160
- messages,
161
- tokenize=True,
162
- add_generation_prompt=True,
163
- return_tensors="pt",
164
- enable_thinking=True
165
- )
166
- ```
167
-
168
- ## Model Capabilities
169
-
170
- ### Mathematical Reasoning
171
- - Complex mathematical problem-solving
172
- - Step-by-step mathematical proofs
173
- - Statistical analysis and computation
174
- - Algebraic and geometric reasoning
175
- - Calculus and advanced mathematics
176
-
177
- ### Programming
178
- - Code generation in multiple languages
179
- - Debugging and code optimization
180
- - Algorithm design and implementation
181
- - Software architecture and design
182
- - Full-stack development assistance
183
-
184
- ### Creative Writing
185
- - Story generation and creative content
186
- - Technical writing and documentation
187
- - Poetry and artistic expression
188
- - Content creation and editing
189
- - Narrative development
190
-
191
- ### Problem Solving
192
- - Logical reasoning and analysis
193
- - Critical thinking and evaluation
194
- - Strategic planning and decision-making
195
- - Complex problem decomposition
196
- - Solution optimization
197
-
198
- ### Agent Capabilities
199
- - Tool use and function calling
200
- - Multi-step reasoning
201
- - Task planning and execution
202
- - Context-aware responses
203
- - Autonomous problem-solving
204
-
205
- ## Technical Specifications
206
-
207
- ### Architecture Details
208
- - **Model Type**: EpicBrelloV1ForCausalLM
209
- - **Attention Mechanism**: Grouped Query Attention (GQA)
210
- - **Position Embedding**: Dynamic RoPE with scaling
211
- - **Normalization**: RMSNorm with epsilon 1e-05
212
- - **Activation**: SiLU (Swish)
213
- - **Quantization**: BF16 precision
214
- - **Optimization**: Flash Attention support
215
-
216
- ### Memory and Performance
217
- - **Model Size**: 1.8B parameters
218
- - **Memory Usage**: ~3.6GB (BF16)
219
- - **Inference Speed**: Optimized for real-time interaction
220
- - **Context Handling**: Up to 256K tokens
221
- - **Batch Processing**: Supports efficient batching
222
-
223
- ### Training and Fine-tuning
224
- - **Base Training**: Tencent Hunyuan foundation
225
- - **Instruction Tuning**: Enhanced for reasoning tasks
226
- - **Chain-of-Thought**: Built-in reasoning capabilities
227
- - **Multi-task Learning**: Optimized for diverse applications
228
-
229
- ## Deployment
230
-
231
- ### Local Deployment
232
- ```bash
233
- # Clone the repository
234
- git clone https://huggingface.co/BrelloES/brello-thinking
235
-
236
- # Install dependencies
237
- pip install transformers torch accelerate
238
-
239
- # Load and run the model
240
- python run_brello.py
241
- ```
242
-
243
- ### Cloud Deployment
244
- ```python
245
- # Using Hugging Face Inference API
246
- from huggingface_hub import InferenceClient
247
-
248
- client = InferenceClient("BrelloES/brello-thinking")
249
- response = client.text_generation("What is 2+2?", max_new_tokens=100)
250
- print(response)
251
- ```
252
-
253
- ### Docker Deployment
254
- ```dockerfile
255
- FROM pytorch/pytorch:latest
256
- RUN pip install transformers accelerate
257
- COPY . /app
258
- WORKDIR /app
259
- CMD ["python", "run_brello.py"]
260
- ```
261
-
262
- ## License
263
-
264
- This model is proprietary software created by Epic Systems and engineered by Rehan Temkar. All rights reserved.
265
-
266
- ## Contact
267
-
268
- - **Creator**: Epic Systems
269
- - **Engineer**: Rehan Temkar
270
- - **Model**: Brello Thinking v1.0.0
271
- - **Repository**: https://huggingface.co/BrelloES/brello-thinking
272
-
273
- ---
274
-
275
- *Brello Thinking - Advanced AI Reasoning by Epic Systems*