| Step | \n", "Training Loss | \n", "
|---|---|
| 1 | \n", "11.384400 | \n", "
" ] }, "metadata": {} }, { "output_type": "execute_result", "data": { "text/plain": [ "('./peft-dialogue-summary-checkpoint-local/tokenizer_config.json',\n", " './peft-dialogue-summary-checkpoint-local/special_tokens_map.json',\n", " './peft-dialogue-summary-checkpoint-local/vocab.json',\n", " './peft-dialogue-summary-checkpoint-local/merges.txt',\n", " './peft-dialogue-summary-checkpoint-local/added_tokens.json',\n", " './peft-dialogue-summary-checkpoint-local/tokenizer.json')" ] }, "metadata": {}, "execution_count": 23 } ] }, { "cell_type": "code", "source": [ "peft_model = PeftModel.from_pretrained(model,\n", " './peft-dialogue-summary-checkpoint-local',\n", " torch_dtype=torch.bfloat16,\n", " is_trainable=False)\n", "peft_model" ], "metadata": { "id": "NyMm1-INNQ5s", "colab": { "base_uri": "https://localhost:8080/" }, "outputId": "62e22870-5389-4f2b-d922-0c0b2d6144fe" }, "execution_count": 24, "outputs": [ { "output_type": "execute_result", "data": { "text/plain": [ "PeftModelForSeq2SeqLM(\n", " (base_model): LoraModel(\n", " (model): BartForConditionalGeneration(\n", " (model): BartModel(\n", " (shared): Embedding(50264, 1024, padding_idx=1)\n", " (encoder): BartEncoder(\n", " (embed_tokens): Embedding(50264, 1024, padding_idx=1)\n", " (embed_positions): BartLearnedPositionalEmbedding(1026, 1024)\n", " (layers): ModuleList(\n", " (0-11): 12 x BartEncoderLayer(\n", " (self_attn): BartAttention(\n", " (k_proj): Linear(in_features=1024, out_features=1024, bias=True)\n", " (v_proj): Linear(\n", " in_features=1024, out_features=1024, bias=True\n", " (lora_dropout): ModuleDict(\n", " (default): Dropout(p=0.05, inplace=False)\n", " )\n", " (lora_A): ModuleDict(\n", " (default): Linear(in_features=1024, out_features=8, bias=False)\n", " )\n", " (lora_B): ModuleDict(\n", " (default): Linear(in_features=8, out_features=1024, bias=False)\n", " )\n", " (lora_embedding_A): ParameterDict()\n", " (lora_embedding_B): ParameterDict()\n", " )\n", " (q_proj): Linear(\n", " in_features=1024, out_features=1024, bias=True\n", " (lora_dropout): ModuleDict(\n", " (default): Dropout(p=0.05, inplace=False)\n", " )\n", " (lora_A): ModuleDict(\n", " (default): Linear(in_features=1024, out_features=8, bias=False)\n", " )\n", " (lora_B): ModuleDict(\n", " (default): Linear(in_features=8, out_features=1024, bias=False)\n", " )\n", " (lora_embedding_A): ParameterDict()\n", " (lora_embedding_B): ParameterDict()\n", " )\n", " (out_proj): Linear(in_features=1024, out_features=1024, bias=True)\n", " )\n", " (self_attn_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n", " (activation_fn): GELUActivation()\n", " (fc1): Linear(in_features=1024, out_features=4096, bias=True)\n", " (fc2): Linear(in_features=4096, out_features=1024, bias=True)\n", " (final_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n", " )\n", " )\n", " (layernorm_embedding): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n", " )\n", " (decoder): BartDecoder(\n", " (embed_tokens): Embedding(50264, 1024, padding_idx=1)\n", " (embed_positions): BartLearnedPositionalEmbedding(1026, 1024)\n", " (layers): ModuleList(\n", " (0-11): 12 x BartDecoderLayer(\n", " (self_attn): BartAttention(\n", " (k_proj): Linear(in_features=1024, out_features=1024, bias=True)\n", " (v_proj): Linear(\n", " in_features=1024, out_features=1024, bias=True\n", " (lora_dropout): ModuleDict(\n", " (default): Dropout(p=0.05, inplace=False)\n", " )\n", " (lora_A): ModuleDict(\n", " (default): Linear(in_features=1024, out_features=8, bias=False)\n", " )\n", " (lora_B): ModuleDict(\n", " (default): Linear(in_features=8, out_features=1024, bias=False)\n", " )\n", " (lora_embedding_A): ParameterDict()\n", " (lora_embedding_B): ParameterDict()\n", " )\n", " (q_proj): Linear(\n", " in_features=1024, out_features=1024, bias=True\n", " (lora_dropout): ModuleDict(\n", " (default): Dropout(p=0.05, inplace=False)\n", " )\n", " (lora_A): ModuleDict(\n", " (default): Linear(in_features=1024, out_features=8, bias=False)\n", " )\n", " (lora_B): ModuleDict(\n", " (default): Linear(in_features=8, out_features=1024, bias=False)\n", " )\n", " (lora_embedding_A): ParameterDict()\n", " (lora_embedding_B): ParameterDict()\n", " )\n", " (out_proj): Linear(in_features=1024, out_features=1024, bias=True)\n", " )\n", " (activation_fn): GELUActivation()\n", " (self_attn_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n", " (encoder_attn): BartAttention(\n", " (k_proj): Linear(in_features=1024, out_features=1024, bias=True)\n", " (v_proj): Linear(\n", " in_features=1024, out_features=1024, bias=True\n", " (lora_dropout): ModuleDict(\n", " (default): Dropout(p=0.05, inplace=False)\n", " )\n", " (lora_A): ModuleDict(\n", " (default): Linear(in_features=1024, out_features=8, bias=False)\n", " )\n", " (lora_B): ModuleDict(\n", " (default): Linear(in_features=8, out_features=1024, bias=False)\n", " )\n", " (lora_embedding_A): ParameterDict()\n", " (lora_embedding_B): ParameterDict()\n", " )\n", " (q_proj): Linear(\n", " in_features=1024, out_features=1024, bias=True\n", " (lora_dropout): ModuleDict(\n", " (default): Dropout(p=0.05, inplace=False)\n", " )\n", " (lora_A): ModuleDict(\n", " (default): Linear(in_features=1024, out_features=8, bias=False)\n", " )\n", " (lora_B): ModuleDict(\n", " (default): Linear(in_features=8, out_features=1024, bias=False)\n", " )\n", " (lora_embedding_A): ParameterDict()\n", " (lora_embedding_B): ParameterDict()\n", " )\n", " (out_proj): Linear(in_features=1024, out_features=1024, bias=True)\n", " )\n", " (encoder_attn_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n", " (fc1): Linear(in_features=1024, out_features=4096, bias=True)\n", " (fc2): Linear(in_features=4096, out_features=1024, bias=True)\n", " (final_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n", " )\n", " )\n", " (layernorm_embedding): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n", " )\n", " )\n", " (lm_head): Linear(in_features=1024, out_features=50264, bias=False)\n", " )\n", " )\n", ")" ] }, "metadata": {}, "execution_count": 24 } ] }, { "cell_type": "code", "source": [ "print(trainable_model_params(model))" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "bAgC8ajWNx-R", "outputId": "8085d9a7-9c41-4b94-e149-632d680a45cb" }, "execution_count": 25, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "trainable params: 73728\n", "all params: 407470080\n", "percentage of trainable params: 0.018094089264173704\n" ] } ] }, { "cell_type": "code", "source": [ "test_index=30\n", "dialogue=dataset['test'][test_index]['dialogue']\n", "summary=dataset['test'][test_index]['summary']\n", "prompt=f\"\"\"\n", "Summarize the conversation:\n", "{dialogue}\n", "\n", "Summary:\n", "\"\"\"\n", "inputs=tokenizer(prompt, return_tensors='pt')\n", "peft_output=tokenizer.decode(\n", " peft_model.generate(\n", " input_ids=inputs['input_ids'].cuda(),\n", " generation_config=GenerationConfig(max_new_tokens=200, temperatue=1, do_sample=True),\n", " )[0],\n", " skip_special_tokens=True\n", ")\n", "original_output=tokenizer.decode(\n", " model.generate(\n", " input_ids=inputs['input_ids'].cuda(),\n", " generation_config=GenerationConfig(max_new_tokens=200,temperature=1, do_sample=True),\n", " )[0],\n", " skip_special_tokens=True\n", ")\n", "print(prompt)\n", "print(dash_line)\n", "print('BASELINE HUMAN SUMMARY:')\n", "print(summary)\n", "print(dash_line)\n", "print('ORIGINAL MODEL SUMMARY:')\n", "print(original_output)\n", "print(dash_line)\n", "print('PEFT MODEL SUMMARY')\n", "print(peft_output)" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "RbfJyX2uOMPH", "outputId": "ba6f3196-0cb1-448d-fdd2-c3cd409b4de3" }, "execution_count": 27, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "\n", "Summarize the conversation:\n", "#Person1#: Where are you going for your trip?\n", "#Person2#: I think Hebei is a good place.\n", "#Person1#: But I heard the north of China are experiencing severe sandstorms!\n", "#Person2#: Really?\n", "#Person1#: Yes, it's said that Hebes was experiencing six degree strong winds.\n", "#Person2#: How do these storms affect the people who live in these areas?\n", "#Person1#: The report said the number of people with respiratory tract infections tended to rise after sandstorms. The sand gets into people's noses and throats and creates irritation.\n", "#Person2#: It sounds that sandstorms are trouble for everybody!\n", "#Person1#: You are quite right.\n", "\n", "Summary:\n", "\n", "---------------------------------------------------------------------------------------------------\n", "BASELINE HUMAN SUMMARY:\n", "#Person2# plans to have a trip in Hebei but #Person1# says there are sandstorms in there.\n", "---------------------------------------------------------------------------------------------------\n", "ORIGINAL MODEL SUMMARY:\n", "TheIs based on interviews with CNN. See CNN.com.\n", "---------------------------------------------------------------------------------------------------\n", "PEFT MODEL SUMMARY\n", "Reality TV chat shows focus on the situation in China, often focusing on the difficulties that the country encounters. The Chinese are still reeling from the impact of typhoons and sandstorms.\n" ] } ] } ] }