Updated notebooks wip

sia/llm_engine/qwq_llm_engine.ipynb

@@ -57,7 +57,6 @@
     "    load_in_4bit = True, # False for LoRA 16bit\n",
     "    fast_inference = True, # Enable vLLM fast inference\n",
     "    gpu_memory_utilization = 0.8, # Reduce if out of memory\n",
-    "    tokenizer = tokenizer,\n",
     ")"
    ]
   },
@@ -175,19 +174,7 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [
+   "outputs": [],
-    {
-     "ename": "",
-     "evalue": "",
-     "output_type": "error",
-     "traceback": [
-      "\u001b[1;31mThe Kernel crashed while executing code in the current cell or a previous cell. \n",
-      "\u001b[1;31mPlease review the code in the cell(s) to identify a possible cause of the failure. \n",
-      "\u001b[1;31mClick <a href='https://aka.ms/vscodeJupyterKernelCrash'>here</a> for more info. \n",
-      "\u001b[1;31mView Jupyter <a href='command:jupyter.viewOutput'>log</a> for further details."
-     ]
-    }
-   ],
    "source": [
     "\n",
     "generation_thread.join()"
@@ -198,7 +185,7 @@
   "kernelspec": {
    "display_name": "sia",
    "language": "python",
-   "name": "python3"
+   "name": "sia"
   },
   "language_info": {
    "codemirror_mode": {

tools/train/train/qwq.ipynb

@@ -54,16 +54,6 @@
     "dataset.validate()"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "max_seq_length = 2048 # Can increase for longer reasoning traces\n",
-    "lora_rank = 64 # Larger rank = smarter, but slower"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -87,12 +77,7 @@
    "source": [
     "model, _returned_tokenizer = FastLanguageModel.from_pretrained(\n",
     "    model_name = args.base_model,\n",
-    "    max_seq_length = max_seq_length,\n",
-    "    load_in_4bit = True, # False for LoRA 16bit\n",
-    "    fast_inference = True, # Enable vLLM fast inference\n",
-    "    max_lora_rank = lora_rank,\n",
     "    gpu_memory_utilization = 0.5, # Reduce if out of memory\n",
-    "    tokenizer = tokenizer,\n",
     ")"
    ]
   },
@@ -104,18 +89,14 @@
    "source": [
     "model = FastLanguageModel.get_peft_model(\n",
     "    model,\n",
-    "    r = lora_rank, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128\n",
     "    target_modules = [\n",
     "        \"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\",\n",
     "        \"gate_proj\", \"up_proj\", \"down_proj\",\n",
-    "    ], # Remove QKVO if out of memory\n",
+    "    ],\n",
-    "    lora_alpha = lora_rank,\n",
     "    lora_dropout = 0, # Supports any, but = 0 is optimized\n",
     "    bias = \"none\",    # Supports any, but = \"none\" is optimized\n",
     "    use_gradient_checkpointing = \"unsloth\", # True or \"unsloth\" for very long context\n",
     "    random_state = 3407,\n",
-    "    use_rslora = False,  # We support rank stabilized LoRA\n",
-    "    loftq_config = None, # And LoftQ\n",
     ")"
    ]
   },
@@ -157,7 +138,6 @@
     "    args=training_args,\n",
     "    train_dataset=dataset.to_transformers_dataset(tokenizer),\n",
     "    dataset_text_field=\"messages\",\n",
-    "    max_seq_length=max_seq_length,\n",
     ")"
    ]
   },
@@ -179,16 +159,16 @@
     "model.save_pretrained_merged(\n",
     "    str(args.output_dir), \n",
     "    tokenizer=tokenizer,\n",
-    "    #save_method=\"merged_4bit_forced\"\n",
+    "    save_method=\"merged_4bit_forced\"\n",
     ")"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3",
+   "display_name": "train",
    "language": "python",
-   "name": "python3"
+   "name": "train"
   },
   "language_info": {
    "codemirror_mode": {