Update README.md
Browse files
README.md
CHANGED
|
@@ -9,12 +9,30 @@ base_model: MiniMaxAI/MiniMax-M2
|
|
| 9 |
|
| 10 |
## Model Details
|
| 11 |
|
|
|
|
|
|
|
| 12 |
- **Quantization Method:** cyankiwi AWQ v1.0
|
| 13 |
- **Bits:** 4
|
| 14 |
- **Group Size:** 32
|
| 15 |
- **Calibration Dataset:** [nvidia/Llama-Nemotron-Post-Training-Dataset](https://huggingface.co/datasets/nvidia/Llama-Nemotron-Post-Training-Dataset)
|
| 16 |
- **Quantization Tool:** [llm-compressor](https://github.com/vllm-project/llm-compressor)
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
## Inference
|
| 19 |
|
| 20 |
### Prerequisite
|
|
|
|
| 9 |
|
| 10 |
## Model Details
|
| 11 |
|
| 12 |
+
### Quantization Details
|
| 13 |
+
|
| 14 |
- **Quantization Method:** cyankiwi AWQ v1.0
|
| 15 |
- **Bits:** 4
|
| 16 |
- **Group Size:** 32
|
| 17 |
- **Calibration Dataset:** [nvidia/Llama-Nemotron-Post-Training-Dataset](https://huggingface.co/datasets/nvidia/Llama-Nemotron-Post-Training-Dataset)
|
| 18 |
- **Quantization Tool:** [llm-compressor](https://github.com/vllm-project/llm-compressor)
|
| 19 |
|
| 20 |
+
### Memory Usage
|
| 21 |
+
|
| 22 |
+
| **Type** | **MiniMax-M2** | **MiniMax-M2-AWQ-4bit** |
|
| 23 |
+
|:---------------:|:----------------:|:----------------:|
|
| 24 |
+
| **Memory Size** | 214.3 GB | 121.5 GB |
|
| 25 |
+
| **KV Cache per Token** | 124.0 kB | 31.0 kB |
|
| 26 |
+
| **KV Cache per Context** | 23.3 GB | 5.8 GB |
|
| 27 |
+
|
| 28 |
+
### Evaluations
|
| 29 |
+
|
| 30 |
+
| **Benchmarks** | **MiniMax-M2** | **MiniMax-M2-AWQ-4bit** |
|
| 31 |
+
|:---------------:|:----------------:|:----------------:|
|
| 32 |
+
| **Perplexity** | 1.54984 | 1.54743 |
|
| 33 |
+
|
| 34 |
+
- **Evaluation Context Length:** 16384
|
| 35 |
+
|
| 36 |
## Inference
|
| 37 |
|
| 38 |
### Prerequisite
|