Update README.md
Browse files
README.md
CHANGED
|
@@ -1,15 +1,12 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
base_model: Qwen/Qwen3-Embedding-4B
|
| 4 |
-
base_model_relation: merge
|
| 5 |
datasets:
|
| 6 |
- HIT-TMG/JevEmbed-Data
|
| 7 |
pipeline_tag: feature-extraction
|
| 8 |
library_name: sentence-transformers
|
| 9 |
tags:
|
| 10 |
-
- sentence-transformers
|
| 11 |
- jevembed
|
| 12 |
-
- lora
|
| 13 |
---
|
| 14 |
|
| 15 |
# JevEmbed-Qwen3-Embedding-4B
|
|
@@ -44,4 +41,4 @@ Qwen3-Embedding-4B and the final JevEmbed LoRA checkpoint were evaluated on all
|
|
| 44 |
|
| 45 |
Training ran for one epoch on the 1,601,157-question JevEmbed-Data training split. The final checkpoint is step **3,127**. Training used 16 GPUs across four nodes, per-GPU batch 4, gradient accumulation 8 (effective batch 512), BF16, LoRA rank 64, alpha 32, dropout 0.05, and Q/K/V projection targets. The learning rate was 2 × 10⁻⁴ with 10% warmup. Inputs were truncated at 1,024 tokens. The training objective is described in the [JevEmbed implementation](https://github.com/HITsz-TMG/JevEmbed/blob/main/src/jevembed/training/objective.py).
|
| 46 |
|
| 47 |
-
The source Qwen model is Apache 2.0 licensed. Training-data source licenses vary; see the dataset's [source report](https://huggingface.co/datasets/HIT-TMG/JevEmbed-Data/blob/main/docs/PROCESSING_REPORT.md).
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
base_model: Qwen/Qwen3-Embedding-4B
|
|
|
|
| 4 |
datasets:
|
| 5 |
- HIT-TMG/JevEmbed-Data
|
| 6 |
pipeline_tag: feature-extraction
|
| 7 |
library_name: sentence-transformers
|
| 8 |
tags:
|
|
|
|
| 9 |
- jevembed
|
|
|
|
| 10 |
---
|
| 11 |
|
| 12 |
# JevEmbed-Qwen3-Embedding-4B
|
|
|
|
| 41 |
|
| 42 |
Training ran for one epoch on the 1,601,157-question JevEmbed-Data training split. The final checkpoint is step **3,127**. Training used 16 GPUs across four nodes, per-GPU batch 4, gradient accumulation 8 (effective batch 512), BF16, LoRA rank 64, alpha 32, dropout 0.05, and Q/K/V projection targets. The learning rate was 2 × 10⁻⁴ with 10% warmup. Inputs were truncated at 1,024 tokens. The training objective is described in the [JevEmbed implementation](https://github.com/HITsz-TMG/JevEmbed/blob/main/src/jevembed/training/objective.py).
|
| 43 |
|
| 44 |
+
The source Qwen model is Apache 2.0 licensed. Training-data source licenses vary; see the dataset's [source report](https://huggingface.co/datasets/HIT-TMG/JevEmbed-Data/blob/main/docs/PROCESSING_REPORT.md).
|