Add training callback to send predictions to WandB table

.vscode/launch.json

@@ -0,0 +1,21 @@
+{
+    // Use IntelliSense to learn about possible attributes.
+    // Hover to view descriptions of existing attributes.
+    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "train",
+            "type": "python",
+            "request": "launch",
+            "module": "accelerate.commands.launch",
+            "args": [
+                "${workspaceFolder}/scripts/finetune.py",
+                // "${file}",
+                "${workspaceFolder}/examples/llama-2/tiny-random.yml",
+            ], // other args comes after train.py
+            "console": "integratedTerminal",
+            // "env": {"CUDA_LAUNCH_BLOCKING": "1"}
+          },
+    ]
+}

docker-compose.yaml

@@ -1,9 +1,10 @@
 # version: '3.8'
 services:
   axolotl:
-    build:
-      context: .
-      dockerfile: ./docker/Dockerfile
+    # build:
+    #   context: .
+    #   dockerfile: ./docker/Dockerfile
+    image: winglian/axolotl:main-py3.10-cu118-2.0.1
     volumes:
       - .:/workspace/axolotl
       - ~/.cache/huggingface/:/root/.cache/huggingface/
@@ -15,6 +16,6 @@ services:
         reservations:
           devices:
             - driver: nvidia
-              # count: 1
+              count: 1
               capabilities: [gpu]
     command: tail -f /dev/null

examples/llama-2/tiny-random.yml

@@ -0,0 +1,93 @@
+# anushehchaudry/llama-2-tiny-random
+# base_model: anushehchaudry/llama-2-tiny-random
+# base_model_config: anushehchaudry/llama-2-tiny-random
+
+# base_model: JackFram/llama-68m
+# base_model_config: JackFram/llama-68m
+
+base_model: PY007/TinyLlama-1.1B-step-50K-105b
+base_model_config: PY007/TinyLlama-1.1B-step-50K-105b
+
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+
+datasets:
+  # - path: mhenrichsen/alpaca_2k_test
+  #   type: alpaca
+  # - path: teknium/GPT4-LLM-Cleaned
+  #   type: alpaca
+  - path: Glavin001/startup-interviews
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+# val_set_size: 0.01
+val_set_size: 0.1
+# output_dir: ./lora-out
+# output_dir: ./lora-2-out
+output_dir: ./lora-5-out
+
+# sequence_len: 4096
+sequence_len: 2048
+# sequence_len: 256
+# sequence_len: 512
+# sample_packing: true
+sample_packing: false
+
+adapter: lora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project: test-issue-490
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+# micro_batch_size: 2
+micro_batch_size: 16
+# num_epochs: 3
+# num_epochs: 0.001
+# num_epochs: 0.01
+# num_epochs: 1
+num_epochs: 5
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+eval_steps: 10
+# eval_steps: 20
+# eval_steps: 2
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"