End of training

README.md

@@ -22,7 +22,6 @@ adapter: lora
 base_model: huggyllama/llama-7b
 bf16: false
 chat_template: llama3
-dataset_prepared_path: null
 datasets:
 - data_files:
   - e159c7177b77ec6f_train_data.json
@@ -42,8 +41,8 @@ deepspeed: null
 early_stopping_patience: null
 eval_max_new_tokens: 128
 eval_table_size: null
-evals_per_epoch: 4
-flash_attention: true
+evals_per_epoch: 1
+flash_attention: false
 fp16: true
 fsdp: null
 fsdp_config: null
@@ -66,7 +65,7 @@ lora_model_dir: null
 lora_r: 8
 lora_target_linear: true
 lr_scheduler: cosine
-max_steps: 10
+max_steps: 1000
 micro_batch_size: 1
 mlflow_experiment_name: /tmp/e159c7177b77ec6f_train_data.json
 model_type: AutoModelForCausalLM
@@ -77,7 +76,7 @@ pad_to_sequence_len: true
 resume_from_checkpoint: null
 s2_attention: null
 sample_packing: false
-saves_per_epoch: 4
+saves_per_epoch: 1
 sequence_len: 1024
 special_tokens:
   pad_token: </s>
@@ -93,7 +92,7 @@ wandb_name: 22b05afc-5100-4d8e-849b-39843d793e1c
 wandb_project: Gradients-On-Demand
 wandb_run: your_name
 wandb_runid: 22b05afc-5100-4d8e-849b-39843d793e1c
-warmup_steps: 10
+warmup_steps: 0
 weight_decay: 0.0
 xformers_attention: null
 
@@ -105,7 +104,7 @@ xformers_attention: null
 
 This model is a fine-tuned version of [huggyllama/llama-7b](https://huggingface.co/huggyllama/llama-7b) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.8663
+- Loss: 0.3320
 
 ## Model description
 
@@ -132,18 +131,14 @@ The following hyperparameters were used during training:
 - total_train_batch_size: 4
 - optimizer: Use OptimizerNames.ADAMW_HF with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
 - lr_scheduler_type: cosine
-- lr_scheduler_warmup_steps: 10
-- training_steps: 10
+- training_steps: 1000
 - mixed_precision_training: Native AMP
 
 ### Training results
 
 | Training Loss | Epoch  | Step | Validation Loss |
 |:-------------:|:------:|:----:|:---------------:|
-| 1.0893        | 0.0001 | 1    | 1.2124          |
-| 1.2625        | 0.0003 | 3    | 1.1848          |
-| 1.0365        | 0.0006 | 6    | 1.0633          |
-| 0.8883        | 0.0010 | 9    | 0.8663          |
+| 0.4271        | 0.1071 | 1000 | 0.3320          |
 
 
 ### Framework versions

adapter_config.json

@@ -20,13 +20,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
+    "o_proj",
+    "k_proj",
+    "up_proj",
     "q_proj",
-    "gate_proj",
-    "down_proj",
     "v_proj",
-    "up_proj",
-    "o_proj",
-    "k_proj"
+    "gate_proj",
+    "down_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd87ab34ae8a609486260fb38d56115626e57fafc1714b0c2827dc2abbe1cb00
+oid sha256:e729e80eac7ef4f85f592626fc8cbef243c2a43888fe92b292843cf966f7e0a0
 size 80115210

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:adfbbddb6d36c12289912b5fa1c4a076dc9ec72221af86dd87e0083ff3608563
+oid sha256:719df9ddf670b10e0acd25b3b60a7525f8c19f58481945e574721fa5a6ae2056
 size 80013120

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec54d060fa3c10be7a79720000a71081ddc512eda26e7a24166cfc8633b27fc3
+oid sha256:86b878855dd3a962afd8534d40010f416a8fb1bce5a59608028eb56c6136a568
 size 6776