ysr/deepseek-rust-1.3b-lora-32

README.md

@@ -20,7 +20,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [deepseek-ai/deepseek-coder-1.3b-base](https://huggingface.co/deepseek-ai/deepseek-coder-1.3b-base) on the generator dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.4360
+- Loss: 0.4301
 
 ## Model description
 
@@ -39,7 +39,7 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 0.0002
+- learning_rate: 0.0003
 - train_batch_size: 1
 - eval_batch_size: 1
 - seed: 42
@@ -48,27 +48,35 @@ The following hyperparameters were used during training:
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.05
-- lr_scheduler_warmup_steps: 1
+- lr_scheduler_warmup_steps: 20
 - num_epochs: 5
 
 ### Training results
 
 | Training Loss | Epoch | Step | Validation Loss |
 |:-------------:|:-----:|:----:|:---------------:|
-| 0.6641        | 0.59  | 25   | 0.5224          |
-| 0.4747        | 1.18  | 50   | 0.4685          |
-| 0.4392        | 1.77  | 75   | 0.4519          |
-| 0.4378        | 2.37  | 100  | 0.4437          |
-| 0.4189        | 2.96  | 125  | 0.4389          |
-| 0.4209        | 3.55  | 150  | 0.4367          |
-| 0.4201        | 4.14  | 175  | 0.4360          |
-| 0.4163        | 4.73  | 200  | 0.4360          |
+| 0.8205        | 0.3   | 25   | 0.5577          |
+| 0.5013        | 0.59  | 50   | 0.4856          |
+| 0.4588        | 0.89  | 75   | 0.4663          |
+| 0.4503        | 1.18  | 100  | 0.4556          |
+| 0.4319        | 1.48  | 125  | 0.4482          |
+| 0.4293        | 1.77  | 150  | 0.4429          |
+| 0.4272        | 2.07  | 175  | 0.4387          |
+| 0.4153        | 2.37  | 200  | 0.4365          |
+| 0.4111        | 2.66  | 225  | 0.4343          |
+| 0.4031        | 2.96  | 250  | 0.4322          |
+| 0.406         | 3.25  | 275  | 0.4317          |
+| 0.3996        | 3.55  | 300  | 0.4309          |
+| 0.3959        | 3.84  | 325  | 0.4303          |
+| 0.3989        | 4.14  | 350  | 0.4302          |
+| 0.4035        | 4.43  | 375  | 0.4303          |
+| 0.3929        | 4.73  | 400  | 0.4301          |
 
 
 ### Framework versions
 
 - PEFT 0.10.0
-- Transformers 4.39.2
+- Transformers 4.39.3
 - Pytorch 2.2.1
 - Datasets 2.18.0
 - Tokenizers 0.15.2

adapter_config.json

@@ -16,14 +16,14 @@
   "megatron_core": "megatron.core",
   "modules_to_save": null,
   "peft_type": "LORA",
-  "r": 64,
+  "r": 32,
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "v_proj",
-    "o_proj",
+    "q_proj",
     "k_proj",
-    "q_proj"
+    "v_proj",
+    "o_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2fd6460db179ca1736b9c7447f6f1ac3ef8d5ca5bc0c962a0a70f4f375627c71
-size 50357632
+oid sha256:84a723e022024cdfece83514750f0c6efcf4dc33aefc626f199a1553a4a60f8d
+size 25191728

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3d201d74214f993646ae033925c94201c44cfcc9da3d21d28f5ab25ce99f8ac
+oid sha256:2b08e1878e76116da9d989483eca71f784ccf29c23ca5c4fb9e809222fcba214
 size 4920