Andrii Maslovskyi
commited on
Commit
Β·
5d7be8b
1
Parent(s):
95cabdb
Update README with comprehensive system requirements and performance expectations
Browse files- Added detailed memory and storage requirements for optimal performance.
- Included hardware compatibility information across various platforms.
- Documented device categories and their respective performance metrics.
- Specified software dependencies for core and optional packages to aid setup.
README.md
CHANGED
@@ -112,10 +112,54 @@ print(response)
|
|
112 |
|
113 |
### π§ **System Requirements**
|
114 |
|
|
|
115 |
- **Minimum RAM**: 21GB (base model + LoRA adapter + working memory)
|
116 |
-
- **Recommended**: 48GB+ for optimal performance
|
117 |
-
- **
|
118 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
119 |
|
120 |
### π
**Strengths & Use Cases**
|
121 |
|
|
|
112 |
|
113 |
### π§ **System Requirements**
|
114 |
|
115 |
+
#### **πΎ Memory Requirements**
|
116 |
- **Minimum RAM**: 21GB (base model + LoRA adapter + working memory)
|
117 |
+
- **Recommended RAM**: 48GB+ for optimal performance and concurrent operations
|
118 |
+
- **Sweet Spot**: 32GB+ provides excellent performance for most use cases
|
119 |
+
|
120 |
+
#### **πΏ Storage Requirements**
|
121 |
+
- **LoRA Adapter**: 182MB (this model)
|
122 |
+
- **Base Model**: ~16GB (Qwen3-8B, downloaded separately)
|
123 |
+
- **Cache & Dependencies**: ~2-3GB (transformers, tokenizers, PyTorch)
|
124 |
+
- **Total Storage**: ~19GB for complete setup
|
125 |
+
|
126 |
+
#### **π₯οΈ Hardware Compatibility**
|
127 |
+
|
128 |
+
| **Platform** | **Status** | **Performance** | **Notes** |
|
129 |
+
| ---------------------------- | ----------- | ----------------- | ---------------------------- |
|
130 |
+
| **Apple Silicon (M1/M2/M3)** | β
Excellent | Fast inference | CPU-optimized, MPS supported |
|
131 |
+
| **Intel/AMD x86-64** | β
Excellent | Good performance | 16+ cores recommended |
|
132 |
+
| **NVIDIA GPU** | β
Optimal | Fastest inference | RTX 4090/5090, A100, H100 |
|
133 |
+
| **AMD GPU** | β οΈ Limited | Basic support | ROCm required, experimental |
|
134 |
+
|
135 |
+
#### **π± Device Categories**
|
136 |
+
|
137 |
+
| **Device Type** | **RAM** | **Performance** | **Use Case** |
|
138 |
+
| ------------------- | ------- | --------------- | --------------------------- |
|
139 |
+
| **High-end Laptop** | 32-64GB | π’ Excellent | Development, personal use |
|
140 |
+
| **Workstation** | 64GB+ | π’ Optimal | Team deployment, production |
|
141 |
+
| **Cloud Instance** | 32GB+ | π’ Scalable | API serving, multiple users |
|
142 |
+
| **Entry Laptop** | 16-24GB | π‘ Limited | Light testing only |
|
143 |
+
|
144 |
+
#### **β‘ Performance Expectations**
|
145 |
+
|
146 |
+
- **Loading Time**: 30-90 seconds (depending on hardware)
|
147 |
+
- **First Response**: 60-120 seconds (model warming)
|
148 |
+
- **Subsequent Responses**: 30-60 seconds average
|
149 |
+
- **Tokens per Second**: 2-5 tokens/sec (CPU), 10-20 tokens/sec (GPU)
|
150 |
+
|
151 |
+
#### **π§ Software Dependencies**
|
152 |
+
```bash
|
153 |
+
# Core requirements
|
154 |
+
torch>=2.0.0
|
155 |
+
transformers>=4.35.0
|
156 |
+
peft>=0.5.0
|
157 |
+
|
158 |
+
# Optional but recommended
|
159 |
+
accelerate>=0.24.0
|
160 |
+
bitsandbytes>=0.41.0 # For quantization
|
161 |
+
flash-attn>=2.0.0 # For GPU optimization
|
162 |
+
```
|
163 |
|
164 |
### π
**Strengths & Use Cases**
|
165 |
|