End of training

README.md

@@ -4,6 +4,7 @@ library_name: peft
 license: apache-2.0
 tags:
 - axolotl
+- relora
 - generated_from_trainer
 model-index:
 - name: pythia-160m-storytelling
@@ -19,22 +20,37 @@ should probably proofread and complete it, then remove this comment. -->
 axolotl version: `0.4.1`
 ```yaml
 base_model: EleutherAI/pythia-160m-deduped
-load_in_8bit: false
+load_in_8bit: 
 datasets:
   - path: jtatman/storywriting_combined_instruct
     type: alpaca
 dataset_prepared_path: ds-storytelling
-val_set_size: 0.05
+chat_template: inst
+val_set_size: 0.01
 adapter: lora
 lora_model_dir: 
 sequence_len: 2048
 lora_r: 16
-lora_alpha: 64
+lora_alpha: 32
 lora_dropout: 0.05
 lora_target_modules:
   - query_key_value
-lora_target_linear: 
+lora_target_linear: true
 lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
+lora_modules_to_save:
+  - embed_tokens
+  - lm_head
+lora_on_cpu: false
+# ReLoRA configuration
+# # Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
+# relora_steps: # Number of steps per ReLoRA restart
+# relora_warmup_steps: # Number of per-restart warmup steps
+# relora_anneal_steps: # Number of anneal steps for each relora cycle
+# relora_prune_ratio: # threshold for optimizer magnitude when pruning
+# relora_cpu_offload:  # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings
+relora_steps: 200
+relora_warmup_steps: 10
+relora_cpu_offload: false
 wandb_project: pythia
 wandb_entity:
 wandb_watch:
@@ -43,7 +59,7 @@ wandb_log_model:
 output_dir: ./outputs/lora-alpaca-pythia-160m-storytelling
 gradient_accumulation_steps: 16
 micro_batch_size: 1
-num_epochs: 5
+num_epochs: 1
 learning_rate: 0.0006
 lr_scheduler: cosine_with_restarts
 #cosine_min_lr_ratio: 0.1
@@ -58,19 +74,17 @@ xformers_attention: true
 optimizer: paged_adamw_8bit
 gpu_memory_limit: 8GiB
 hub_model_id: jtatman/pythia-160m-storytelling 
-lora_on_cpu: true
-early_stopping_patience: 3
+early_stopping_patience: 2
 #resume_from_checkpoint: outputs/lora-alpaca-pythia-125m/checkpoint-51040
 auto_resume_from_checkpoints: true
 local_rank:
-weight_decay: 0.1
-chat_template: inst
+weight_decay: 0.0
 #evals_per_epoch: 4
-eval_steps: 2000
+eval_steps: 200
 logging_steps: 1
-save_steps: 2000
+save_steps: 200
 save_total_limit: 5
-warmup_steps: 1000
+warmup_steps: 100
 
 ```
 
@@ -80,7 +94,7 @@ warmup_steps: 1000
 
 This model is a fine-tuned version of [EleutherAI/pythia-160m-deduped](https://huggingface.co/EleutherAI/pythia-160m-deduped) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 10.3843
+- Loss: 2.8975
 
 ## Model description
 
@@ -107,16 +121,18 @@ The following hyperparameters were used during training:
 - total_train_batch_size: 16
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine_with_restarts
-- lr_scheduler_warmup_steps: 1000
-- num_epochs: 5
+- lr_scheduler_warmup_steps: 100
+- num_epochs: 1
 
 ### Training results
 
 | Training Loss | Epoch  | Step | Validation Loss |
 |:-------------:|:------:|:----:|:---------------:|
-| 5.4891        | 0.0012 | 1    | 4.5640          |
-| 8.4799        | 2.4467 | 2000 | 9.1436          |
-| 9.9198        | 4.8944 | 4000 | 10.3843         |
+| 5.5185        | 0.0012 | 1    | 4.8333          |
+| 3.7004        | 0.2348 | 200  | 3.2693          |
+| 3.52          | 0.4696 | 400  | 3.3535          |
+| 3.7836        | 0.7043 | 600  | 2.9896          |
+| 3.3058        | 0.9391 | 800  | 2.8975          |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d341e5db83d74ccdb7d452b50b4cd56947a73515c321d0efc022725975d5ad9
-size 4731832
+oid sha256:e44ce263e6fd885f50d82ca515b9325375b43ee36ededb75acf161ce88bc2e41
+size 48

config.json

@@ -22,7 +22,7 @@
   "rotary_emb_base": 10000,
   "rotary_pct": 0.25,
   "tie_word_embeddings": false,
-  "torch_dtype": "float16",
+  "torch_dtype": "bfloat16",
   "transformers_version": "4.41.2",
   "use_cache": false,
   "use_parallel_residual": true,

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "do_sample": true,
+  "eos_token_id": 0,
+  "transformers_version": "4.41.2"
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:727fc3779b3fe223ff519f231c0c6c4a9f9092e972590479a1a23e8a5cb4c7db
+size 324696090