End of training

README.md

@@ -20,7 +20,7 @@ axolotl version: `0.4.1`
 ```yaml
 adapter: lora
 base_model: unsloth/SmolLM-135M-Instruct
-bf16: auto
+bf16: true
 chat_template: llama3
 dataset_prepared_path: null
 datasets:
@@ -41,19 +41,20 @@ deepspeed: null
 early_stopping_patience: null
 eval_max_new_tokens: 128
 eval_table_size: null
-evals_per_epoch: 4
+evals_per_epoch: 1
 flash_attention: false
 fp16: null
 fsdp: null
 fsdp_config: null
 gradient_accumulation_steps: 4
-gradient_checkpointing: false
+gradient_checkpointing: true
+gradient_clipping: 1.0
 group_by_length: false
 hub_model_id: kk-aivio/test
 hub_repo: null
-hub_strategy: checkpoint
+hub_strategy: end
 hub_token: null
-learning_rate: 0.0002
+learning_rate: 5.0e-05
 load_in_4bit: false
 load_in_8bit: false
 local_rank: null
@@ -65,7 +66,7 @@ lora_model_dir: null
 lora_r: 8
 lora_target_linear: true
 lr_scheduler: cosine
-max_steps: 10
+max_steps: 200
 micro_batch_size: 2
 mlflow_experiment_name: /tmp/8f98792a60ccddb6_train_data.json
 model_type: AutoModelForCausalLM
@@ -76,8 +77,8 @@ pad_to_sequence_len: true
 resume_from_checkpoint: null
 s2_attention: null
 sample_packing: false
-saves_per_epoch: 4
-sequence_len: 512
+saves_per_epoch: 1
+sequence_len: 1024
 strict: false
 tf32: false
 tokenizer_type: AutoTokenizer
@@ -90,9 +91,9 @@ wandb_name: ec775fcf-a2f3-4d2e-b927-b54ca72381d3
 wandb_project: Gradients-On-Demand
 wandb_run: your_name
 wandb_runid: ec775fcf-a2f3-4d2e-b927-b54ca72381d3
-warmup_steps: 10
-weight_decay: 0.0
-xformers_attention: null
+warmup_steps: 5
+weight_decay: 0.01
+xformers_attention: true
 
 ```
 
@@ -121,7 +122,7 @@ More information needed
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
-- learning_rate: 0.0002
+- learning_rate: 5e-05
 - train_batch_size: 2
 - eval_batch_size: 2
 - seed: 42
@@ -129,17 +130,14 @@ The following hyperparameters were used during training:
 - total_train_batch_size: 8
 - optimizer: Use OptimizerNames.ADAMW_BNB 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
+- lr_scheduler_warmup_steps: 5
+- training_steps: 200
 
 ### Training results
 
 | Training Loss | Epoch  | Step | Validation Loss |
 |:-------------:|:------:|:----:|:---------------:|
-| 0.0           | 0.0028 | 1    | nan             |
-| 0.0           | 0.0085 | 3    | nan             |
-| 0.0           | 0.0171 | 6    | nan             |
-| 0.0           | 0.0256 | 9    | nan             |
+| 0.0           | 0.5690 | 200  | nan             |
 
 
 ### Framework versions

adapter_config.json

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

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c1cd7365ed353ff72dee783b0380c3fe47edb5c8b70351ee642727ccb6c9326
+oid sha256:e078c1f09f143010b5531d17e6b0bc125e01e78e3fbe602ae00a89b462c7a32a
 size 9919898

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:30b9abb25769ac7a172d7ca24fec4c2e9535e372fe09c5b9db42c635c5304d92
+oid sha256:4f30f5199b856cf4be6434e8e28931fada05775a8edc189335bd3efbdc55172e
 size 9823216

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b6f35945668f0eea961e60501dc805736f3ab3e53ab40b242bbe9677c855ccc
+oid sha256:6a0db34003c8b83cfde6c82effb9b4ff997a103e07023814f6641daae954cd05
 size 6712