Minor Multi-model FT

Author: ashwinprasadme

src/target_tools/ollama/models_info.md

@@ -0,0 +1,45 @@
+# Models
+
+- codellama:7b-python
+- codellama:13b-python
+- codellama:34b-python
+- codellama:7b-instruct
+- codellama:13b-instruct
+- codellama:34b-instruct
+- llama2:7b
+- llama2:13b
+- llama2:70b
+- vicuna:7b
+- vicuna:13b
+- vicuna:33b
+- phind-codellama:34b-v2
+- phind-codellama:34b-python
+- wizardcoder:7b-python
+- wizardcoder:13b-python
+- wizardcoder:34b-python
+- orca2:7b
+- orca2:13b
+- gpt-3.5-turbo
+- gpt-4
+
+# HG
+
+- codellama/CodeLlama-7b-Python-hf
+- codellama/CodeLlama-13b-Python-hf
+- codellama/CodeLlama-34b-Python-hf
+- codellama/CodeLlama-7b-Instruct-hf
+- codellama/CodeLlama-13b-Instruct-hf
+- codellama/CodeLlama-34b-Instruct-hf
+- meta-llama/Llama-2-7b-hf
+- meta-llama/Llama-2-13b-hf
+- meta-llama/Llama-2-70b-hf
+- lmsys/vicuna-7b-v1.5
+- lmsys/vicuna-13b-v1.5
+- lmsys/vicuna-33b-v1.3
+- Phind/Phind-CodeLlama-34B-v2
+- Phind/Phind-CodeLlama-34B-Python-v1
+- WizardLM/WizardCoder-Python-7B-V1.0
+- WizardLM/WizardCoder-Python-13B-V1.0
+- WizardLM/WizardCoder-Python-34B-V1.0
+- microsoft/Orca-2-7b
+- microsoft/Orca-2-13b

src/target_tools/ollama/src/fine_tuning/llama_fine_tuning.ipynb

@@ -59,6 +59,7 @@
     "from datasets import load_dataset\n",
     "from functools import partial\n",
     "import os\n",
+    "import time\n",
     "from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training, AutoPeftModelForCausalLM\n",
     "import torch\n",
     "from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed, Trainer, TrainingArguments, BitsAndBytesConfig, \\\n",
@@ -357,6 +358,35 @@
     "## Load model"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "models_list = [\n",
+    "    \"codellama/CodeLlama-7b-Python-hf\",\n",
+    "    \"codellama/CodeLlama-13b-Python-hf\",\n",
+    "    \"codellama/CodeLlama-34b-Python-hf\",\n",
+    "    \"codellama/CodeLlama-7b-Instruct-hf\",\n",
+    "    \"codellama/CodeLlama-13b-Instruct-hf\",\n",
+    "    \"codellama/CodeLlama-34b-Instruct-hf\",\n",
+    "    \"meta-llama/Llama-2-7b-hf\",\n",
+    "    \"meta-llama/Llama-2-13b-hf\",\n",
+    "    \"meta-llama/Llama-2-70b-hf\",\n",
+    "    \"lmsys/vicuna-7b-v1.5\",\n",
+    "    \"lmsys/vicuna-13b-v1.5\",\n",
+    "    \"lmsys/vicuna-33b-v1.3\",\n",
+    "    \"Phind/Phind-CodeLlama-34B-v2\",\n",
+    "    \"Phind/Phind-CodeLlama-34B-Python-v1\",\n",
+    "    \"WizardLM/WizardCoder-Python-7B-V1.0\",\n",
+    "    \"WizardLM/WizardCoder-Python-13B-V1.0\",\n",
+    "    \"WizardLM/WizardCoder-Python-34B-V1.0\",\n",
+    "    \"microsoft/Orca-2-7b\",\n",
+    "    \"microsoft/Orca-2-13b\",\n",
+    "]"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -365,10 +395,10 @@
    "outputs": [],
    "source": [
     "# Load model from HF with user's token and with bitsandbytes config\n",
-    "models_list = [\"meta-llama/Llama-2-7b-hf\", \"meta-llama/Llama-2-13b-hf\"]\n",
     "output_dir_str = \"/scratch/hpc-prf-hdgen/ashwin/finetuned_models/ft_{model_name}\"\n",
     "output_dir_merged_str = \"/scratch/hpc-prf-hdgen/ashwin/finetuned_models/ft_{model_name}_merged\"\n",
     "\n",
+    "start_time = time.time()\n",
     "for model_name in models_list:\n",
     "    print(f\"Processing Model: {model_name}\")\n",
     "    # model_name = \"meta-llama/Llama-2-7b-hf\" \n",
@@ -396,7 +426,9 @@
     "\n",
     "    # save tokenizer for easy inference\n",
     "    tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
-    "    tokenizer.save_pretrained(output_merged_dir)\n"
+    "    tokenizer.save_pretrained(output_merged_dir)\n",
+    "\n",
+    "print(f\"DONE! Took{time.time()-start_time}\")"
    ]
   }
  ],

src/target_tools/ollama/src/fine_tuning/requirements.txt

@@ -0,0 +1,8 @@
+torch
+accelerate @ git+https://github.com/huggingface/accelerate.git
+bitsandbytes
+datasets
+transformers @ git+https://github.com/huggingface/transformers.git
+peft @ git+https://github.com/huggingface/peft.git
+trl @ git+https://github.com/lvwerra/trl.git
+scipy