mirror of
https://github.com/deepseek-ai/DeepSeek-V3.git
synced 2025-02-23 06:08:58 -05:00
Improve DeepSeek-V3 Weight File Documentation for Clarity and Readability
- Enhanced sentence structure for better clarity and smoother flow. - Adjusted wording and phrasing to improve accuracy and professionalism. - Optimized the organization of information for better readability, especially in the sections related to parameters and technical details. - Refined formatting and sectioning of the documentation for easier navigation and comprehension.
This commit is contained in:
parent
b5d872ead0
commit
b4e06d883e
@ -2,30 +2,30 @@
|
|||||||
|
|
||||||
## New Fields in `config.json`
|
## New Fields in `config.json`
|
||||||
|
|
||||||
- **model_type**: Specifies the model type, which is updated to `deepseek_v3` in this release.
|
- **model_type**: Specifies the model type, which is now set to `deepseek_v3` in this release.
|
||||||
- **num_nextn_predict_layers**: Indicates the number of Multi-Token Prediction (MTP) Modules. The open-sourced V3 weights include **1 MTP Module** .
|
- **num_nextn_predict_layers**: Defines the number of Multi-Token Prediction (MTP) Modules. The open-sourced V3 weights contain **1 MTP Module**.
|
||||||
- **quantization_config**: Describes the configuration for FP8 quantization.
|
- **quantization_config**: Details the configuration for FP8 quantization.
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## Weight Structure Overview
|
## Weight File Structure Overview
|
||||||
|
|
||||||
The DeepSeek-V3 weight file consists of two main components: **Main Model Weights** and **MTP Modules**.
|
The DeepSeek-V3 weight file is divided into two primary components: **Main Model Weights** and **MTP Modules**.
|
||||||
|
|
||||||
### 1. Main Model Weights
|
### 1. Main Model Weights
|
||||||
|
|
||||||
- **Composition**:
|
- **Composition**:
|
||||||
- Input/output embedding layers and a complete set of 61 Transformer hidden layers.
|
- Includes input/output embedding layers and a full set of 61 Transformer hidden layers.
|
||||||
- **Parameter Count**:
|
- **Parameter Count**:
|
||||||
- Total parameters: **671B**
|
- Total parameters: **671B**
|
||||||
- Activation parameters: **36.7B** (including 0.9B for Embedding and 0.9B for the output Head).
|
- Activation parameters: **36.7B** (which includes 0.9B for Embedding and 0.9B for the Output Head).
|
||||||
|
|
||||||
#### Structural Details
|
#### Structural Details
|
||||||
|
|
||||||
- **Embedding Layer**:
|
- **Embedding Layer**:
|
||||||
- `model.embed_tokens.weight`
|
- `model.embed_tokens.weight`
|
||||||
- **Transformer Hidden Layers**:
|
- **Transformer Hidden Layers**:
|
||||||
- `model.layers.0` to `model.layers.60`, totaling `num_hidden_layers` layers.
|
- From `model.layers.0` to `model.layers.60`, which correspond to `num_hidden_layers` layers.
|
||||||
- **Output Layer**:
|
- **Output Layer**:
|
||||||
- `model.norm.weight`
|
- `model.norm.weight`
|
||||||
- `lm_head.weight`
|
- `lm_head.weight`
|
||||||
@ -33,37 +33,37 @@ The DeepSeek-V3 weight file consists of two main components: **Main Model Weight
|
|||||||
### 2. Multi-Token Prediction (MTP) Modules
|
### 2. Multi-Token Prediction (MTP) Modules
|
||||||
|
|
||||||
- **Composition**:
|
- **Composition**:
|
||||||
- Additional MTP Modules defined by the `num_nextn_predict_layers` field. In this model, the value is set to 1.
|
- These modules are determined by the `num_nextn_predict_layers` parameter. In this model, the value is set to 1.
|
||||||
- **Parameter Count**:
|
- **Parameter Count**:
|
||||||
- Parameters: **11.5B unique parameters**, excluding the shared 0.9B Embedding and 0.9B output Head).
|
- Parameters: **11.5B unique parameters** (excluding the shared 0.9B Embedding and 0.9B Output Head).
|
||||||
- Activation parameters: **2.4B** (including the shared 0.9B Embedding and 0.9B output Head).
|
- Activation parameters: **2.4B** (including the shared 0.9B Embedding and 0.9B Output Head).
|
||||||
|
|
||||||
#### Structural Details
|
#### Structural Details
|
||||||
|
|
||||||
- **embed_tokens**: **Shares parameters** with the Embedding layer of the Main Model weights.
|
- **embed_tokens**: **Shares parameters** with the Main Model’s Embedding layer.
|
||||||
- **enorm & hnorm**: RMSNorm parameters required for speculative decoding.
|
- **enorm & hnorm**: RMSNorm parameters used for speculative decoding.
|
||||||
- **eh_proj**: Parameters for dimensionality reduction projection on the norm results.
|
- **eh_proj**: Parameters used for dimensionality reduction of the normalized outputs.
|
||||||
- **Additional Transformer Hidden Layer**:
|
- **Additional Transformer Hidden Layer**:
|
||||||
- `model.layers.61.self_attn & mlp` (structure identical to the Main Model hidden layers).
|
- `model.layers.61.self_attn & mlp` (these are structured the same as the Main Model hidden layers).
|
||||||
- **shared_head**: **Shares parameters** with the output Head of the Main Model weights.
|
- **shared_head**: **Shares parameters** with the Output Head of the Main Model.
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
### Loading Rules
|
### Layer Loading Rules
|
||||||
|
|
||||||
- **Main Model Weights**: Loaded via the `num_hidden_layers` parameter in `config.json`.
|
- **Main Model Weights**: These are loaded according to the `num_hidden_layers` field in `config.json`.
|
||||||
- **MTP Modules**: Loaded via the `num_nextn_predict_layers` parameter, with layer IDs appended immediately after the Main Model hidden layers. For example:
|
- **MTP Modules**: These are loaded using the `num_nextn_predict_layers` field, with MTP layer IDs appended directly after the Main Model’s hidden layers. For example:
|
||||||
- If `num_hidden_layers = 61` and `num_nextn_predict_layers = 1`, the MTP Module's layer ID is `61`.
|
- With `num_hidden_layers = 61` and `num_nextn_predict_layers = 1`, the MTP Module layer ID will be `61`.
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## FP8 Weight Documentation
|
## FP8 Weight Documentation
|
||||||
|
|
||||||
DeepSeek-V3 natively supports FP8 weight format with 128x128 block scaling.
|
DeepSeek-V3 natively supports the FP8 weight format with 128x128 block scaling.
|
||||||
|
|
||||||
### FP8 Configuration
|
### FP8 Configuration
|
||||||
|
|
||||||
The FP8 weight file introduces a `quantization_config` field to describe the quantization method. Below is an example configuration:
|
The FP8 weight file introduces a `quantization_config` field, which defines the quantization method. Below is an example of the configuration:
|
||||||
|
|
||||||
```json
|
```json
|
||||||
"quantization_config": {
|
"quantization_config": {
|
||||||
@ -75,20 +75,18 @@ The FP8 weight file introduces a `quantization_config` field to describe the qua
|
|||||||
```
|
```
|
||||||
|
|
||||||
- **Quantization Format**:
|
- **Quantization Format**:
|
||||||
- Format type: `fp8` and `e4m3` (corresponding to `torch.float8_e4m3fn`).
|
- Format type: `fp8` and `e4m3` (aligned with `torch.float8_e4m3fn`).
|
||||||
- Weight block size: `128x128`.
|
- Weight block size: `128x128`.
|
||||||
- **Activation Quantization Scheme**:
|
- **Activation Quantization Scheme**:
|
||||||
- Utilizes dynamic activation quantization (`dynamic`).
|
- Uses dynamic activation quantization (`dynamic`).
|
||||||
|
|
||||||
### Dequantization Method
|
### Dequantization Method
|
||||||
|
|
||||||
The FP8 weight file includes a `weight_scale_inv` field, which stores the dequantization scale for each weight block.
|
The FP8 weight file includes a `weight_scale_inv` field, which stores the dequantization scale for each weight block.
|
||||||
|
|
||||||
- **Storage Format**: `float32 Tensor`, stored alongside the weight data.
|
- **Storage Format**: Stored as a `float32 Tensor`, alongside the weight data.
|
||||||
- **Dequantization Formula**:
|
- **Dequantization Formula**:
|
||||||
- If the weight block is not aligned to 128, it is zero-padded to 128 before calculating the scale. After quantization, the padded portion is removed.
|
- If a weight block is not aligned to 128, it is zero-padded to 128 before calculating the scale. The padded portion is discarded after quantization.
|
||||||
- The dequantization process is performed as: `(128x128 weight block) * weight_scale_inv`.
|
- Dequantization is performed using the formula: `(128x128 weight block) * weight_scale_inv`.
|
||||||
|
|
||||||
Through dequantization of the FP8 weights, runtime operations enable online quantization at a granularity of `per-token-per-128-channel`.
|
This dequantization process enables runtime operations to apply online quantization on a per-token, per-128-channel basis.
|
||||||
|
|
||||||
---
|
|
||||||
|
Loading…
Reference in New Issue
Block a user