Merge pull request #34 from togethercomputer/update-finetune

update for FT validation support and LoRA FT

openapi.yaml

@@ -370,7 +370,10 @@ paths:
               properties:
                 training_file:
                   type: string
-                  description: File-ID of a file uploaded to the Together API
+                  description: File-ID of a training file uploaded to the Together API
+                validation_file:
+                  type: string
+                  description: File-ID of a validation file uploaded to the Together API
                 model:
                   type: string
                   description: Name of the base model to run fine-tune job on
@@ -382,6 +385,10 @@ paths:
                   type: integer
                   default: 1
                   description: Number of checkpoints to save during fine-tuning
+                n_evals:
+                  type: integer
+                  default: 0
+                  description: Number of evaluations to be run on a given validation set during training
                 batch_size:
                   type: integer
                   default: 32
@@ -397,6 +404,26 @@ paths:
                 wandb_api_key:
                   type: string
                   description: API key for Weights & Biases integration
+                lora:
+                  type: boolean
+                  description: Whether to enable LoRA training. If not provided, full fine-tuning will be applied.
+                lora_r:
+                  type: integer
+                  default: 8
+                  description: Rank for LoRA adapter weights
+                lora_alpha:
+                  type: integer
+                  default: 8
+                  description: The alpha value for LoRA adapter training. 
+                lora_dropout:
+                  type: number
+                  format: float
+                  default: 0.0
+                  description: The dropout probability for Lora layers.
+                lora_trainable_modules:
+                  type: string
+                  default: 'all-linear'
+                  description: A list of LoRA trainable modules, separated by a comma
       responses:
         '200':
           description: Fine-tuning job initiated successfully
@@ -1656,9 +1683,9 @@ components:
           type: string
         model_output_path:
           type: string
-        TrainingFileNumLines:
+        trainingfile_numlines:
           type: integer
-        TrainingFileSize:
+        trainingfile_size:
           type: integer
         created_at:
           type: string
@@ -1668,6 +1695,8 @@ components:
           type: integer
         n_checkpoints:
           type: integer
+        n_evals:
+          type: integer
         batch_size:
           type: integer
         learning_rate:
@@ -1681,7 +1710,9 @@ components:
         lora_alpha:
           type: integer
         lora_dropout:
-          type: integer
+          type: number
+        lora_trainable_modules:
+          type: string
         status:
           $ref: '#/components/schemas/FinetuneJobStatus'
         job_id: