Update README.md
Browse files
README.md
CHANGED
|
@@ -2,7 +2,9 @@
|
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
|
| 5 |
-
|
|
|
|
|
|
|
| 6 |
|
| 7 |
APUS-xDAN-4.0-MOE is a transformer-based decoder-only language model, developed on a vast corpus of data to ensure robust performance.
|
| 8 |
|
|
@@ -17,7 +19,7 @@ APUS-xDAN-4.0-MOE leverages the innovative Mixture of Experts (MoE) architecture
|
|
| 17 |
Through advanced quantization techniques, our open-source version occupies a mere 42GB, making it seamlessly compatible with consumer-grade GPUs like the 4090 and 3090.
|
| 18 |
The following specifications:
|
| 19 |
|
| 20 |
-
- **Parameters:**
|
| 21 |
- **Architecture:** Mixture of 4 Experts (MoE)
|
| 22 |
- **Experts Utilization:** 2 experts used per token
|
| 23 |
- **Layers:** 60
|
|
@@ -26,7 +28,7 @@ The following specifications:
|
|
| 26 |
- **Additional Features:**
|
| 27 |
- Rotary embeddings (RoPE)
|
| 28 |
- Supports activation sharding and 1.5bit~4bit quantization
|
| 29 |
-
- **Maximum Sequence Length (context):** 32,768 tokens
|
| 30 |
## Usage
|
| 31 |
|
| 32 |
### Initial
|
|
@@ -38,7 +40,7 @@ make LLAMA_CUDA=1
|
|
| 38 |
### Interactive Chat
|
| 39 |
```python
|
| 40 |
|
| 41 |
-
./main -m xDAN-
|
| 42 |
--prompt "You are a helpful assistant." --chatml \
|
| 43 |
--interactive \
|
| 44 |
--temp 0.7 \
|
|
|
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
## Introduction
|
| 8 |
|
| 9 |
APUS-xDAN-4.0-MOE is a transformer-based decoder-only language model, developed on a vast corpus of data to ensure robust performance.
|
| 10 |
|
|
|
|
| 19 |
Through advanced quantization techniques, our open-source version occupies a mere 42GB, making it seamlessly compatible with consumer-grade GPUs like the 4090 and 3090.
|
| 20 |
The following specifications:
|
| 21 |
|
| 22 |
+
- **Parameters:** 136B
|
| 23 |
- **Architecture:** Mixture of 4 Experts (MoE)
|
| 24 |
- **Experts Utilization:** 2 experts used per token
|
| 25 |
- **Layers:** 60
|
|
|
|
| 28 |
- **Additional Features:**
|
| 29 |
- Rotary embeddings (RoPE)
|
| 30 |
- Supports activation sharding and 1.5bit~4bit quantization
|
| 31 |
+
- **Maximum Sequence Length (context):** 32,768 tokens
|
| 32 |
## Usage
|
| 33 |
|
| 34 |
### Initial
|
|
|
|
| 40 |
### Interactive Chat
|
| 41 |
```python
|
| 42 |
|
| 43 |
+
./main -m APUS-xDAN-4.0-quanzied_version.gguf \
|
| 44 |
--prompt "You are a helpful assistant." --chatml \
|
| 45 |
--interactive \
|
| 46 |
--temp 0.7 \
|