update readme and add config file

Author: wjayesh

gamesense/README.md

@@ -46,6 +46,16 @@ GameSense leverages Parameter-Efficient Fine-Tuning (PEFT) techniques to customi
 - Python 3.8+
 - GPU with at least 24GB VRAM (for full model training)
 - ZenML installed and configured
+- Neptune.ai account for experiment tracking
+
+### Environment Setup
+
+1. Set up your Neptune.ai credentials:
+   ```bash
+   # Set your Neptune project name and API token as environment variables
+   export NEPTUNE_PROJECT="your-neptune-workspace/your-project-name"
+   export NEPTUNE_API_TOKEN="your-neptune-api-token"
+   ```
 
 ### Quick Setup
 

gamesense/configs/phi3.5_finetune_cpu.yaml

@@ -0,0 +1,85 @@
+# Apache Software License 2.0
+# 
+# Copyright (c) ZenML GmbH 2024. All rights reserved.
+# 
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+# 
+# http://www.apache.org/licenses/LICENSE-2.0
+# 
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# 
+
+model:
+  name: llm-peft-phi-3.5-mini-instruct-cpu
+  description: "Fine-tune Phi-3.5-mini-instruct on CPU."
+  tags:
+    - llm
+    - peft
+    - phi-3.5
+    - cpu
+  version: 100_steps
+
+settings:
+  docker:
+    parent_image: pytorch/pytorch:2.2.2-runtime
+    requirements: requirements.txt
+    python_package_installer: uv
+    python_package_installer_args:
+      system: null
+    apt_packages: 
+      - git
+    environment:
+      MKL_SERVICE_FORCE_INTEL: "1"
+      # Explicitly disable MPS
+      PYTORCH_ENABLE_MPS_FALLBACK: "0"
+      PYTORCH_MPS_HIGH_WATERMARK_RATIO: "0.0"
+
+parameters:
+  # Uses a smaller model for CPU training
+  base_model_id: microsoft/Phi-3.5-mini-instruct
+  use_fast: False
+  load_in_4bit: False
+  load_in_8bit: False
+  cpu_only: True  # Enable CPU-only mode
+  # Extra conservative dataset size for CPU
+  max_train_samples: 50
+  max_val_samples: 10
+  max_test_samples: 5
+  system_prompt: |
+      Given a target sentence construct the underlying meaning representation of the input sentence as a single function with attributes and attribute values.
+      This function should describe the target string accurately and the function must be one of the following ['inform', 'request', 'give_opinion', 'confirm', 'verify_attribute', 'suggest', 'request_explanation', 'recommend', 'request_attribute'].
+      The attributes must be one of the following: ['name', 'exp_release_date', 'release_year', 'developer', 'esrb', 'rating', 'genres', 'player_perspective', 'has_multiplayer', 'platforms', 'available_on_steam', 'has_linux_release', 'has_mac_release', 'specifier']
+      
+
+steps:
+  prepare_data:
+    parameters:
+      dataset_name: gem/viggo
+      # These settings are now defined at the pipeline level
+      # max_train_samples: 100
+      # max_val_samples: 20
+      # max_test_samples: 10
+
+  finetune:
+    parameters:
+      max_steps: 25  # Further reduced steps for CPU training
+      eval_steps: 5  # More frequent evaluation
+      bf16: False  # Disable bf16 for CPU compatibility
+      per_device_train_batch_size: 1  # Smallest batch size for CPU
+      gradient_accumulation_steps: 2  # Reduced for CPU
+      optimizer: "adamw_torch"  # Use standard AdamW rather than 8-bit for CPU
+      logging_steps: 2  # More frequent logging
+      save_steps: 25  # Save less frequently 
+      save_total_limit: 1  # Keep only the best model
+      evaluation_strategy: "steps"
+
+  promote:
+    parameters:
+      metric: rouge2
+      target_stage: staging