remove handler

4-bit/config.json

@@ -0,0 +1,85 @@
+{
+  "_name_or_path": "bart-large-cnn",
+  "_num_labels": 3,
+  "activation_dropout": 0.0,
+  "activation_function": "gelu",
+  "add_final_layer_norm": false,
+  "architectures": [
+    "BartForConditionalGeneration"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "classif_dropout": 0.0,
+  "classifier_dropout": 0.0,
+  "d_model": 1024,
+  "decoder_attention_heads": 16,
+  "decoder_ffn_dim": 4096,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 12,
+  "decoder_start_token_id": 2,
+  "dropout": 0.1,
+  "early_stopping": true,
+  "encoder_attention_heads": 16,
+  "encoder_ffn_dim": 4096,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 12,
+  "eos_token_id": 2,
+  "force_bos_token_to_be_generated": true,
+  "forced_bos_token_id": 0,
+  "forced_eos_token_id": 2,
+  "gradient_checkpointing": false,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "length_penalty": 2.0,
+  "max_length": 256,
+  "max_position_embeddings": 1024,
+  "min_length": 56,
+  "model_type": "bart",
+  "no_repeat_ngram_size": 3,
+  "normalize_before": false,
+  "num_beams": 4,
+  "num_hidden_layers": 12,
+  "output_past": true,
+  "pad_token_id": 1,
+  "prefix": " ",
+  "quantization_config": {
+    "_load_in_4bit": true,
+    "_load_in_8bit": false,
+    "bnb_4bit_compute_dtype": "bfloat16",
+    "bnb_4bit_quant_storage": "uint8",
+    "bnb_4bit_quant_type": "fp4",
+    "bnb_4bit_use_double_quant": false,
+    "llm_int8_enable_fp32_cpu_offload": false,
+    "llm_int8_has_fp16_weight": false,
+    "llm_int8_skip_modules": null,
+    "llm_int8_threshold": 6.0,
+    "load_in_4bit": true,
+    "load_in_8bit": false,
+    "quant_method": "bitsandbytes"
+  },
+  "scale_embedding": false,
+  "task_specific_params": {
+    "summarization": {
+      "early_stopping": true,
+      "length_penalty": 2.0,
+      "max_length": 142,
+      "min_length": 56,
+      "no_repeat_ngram_size": 3,
+      "num_beams": 4
+    }
+  },
+  "torch_dtype": "float16",
+  "transformers_version": "4.41.2",
+  "use_cache": true,
+  "vocab_size": 50264
+}

4-bit/generation_config.json

@@ -0,0 +1,16 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "decoder_start_token_id": 2,
+  "early_stopping": true,
+  "eos_token_id": 2,
+  "forced_bos_token_id": 0,
+  "forced_eos_token_id": 2,
+  "length_penalty": 2.0,
+  "max_length": 256,
+  "min_length": 56,
+  "no_repeat_ngram_size": 3,
+  "num_beams": 4,
+  "pad_token_id": 1,
+  "transformers_version": "4.41.2"
+}

4-bit/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadcc379c74fc48b0b2395f42794326901c92f98902ced7ebb9c7d369d486db5
+size 306379320

handler.py

@@ -0,0 +1,43 @@
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, BitsAndBytesConfig
+from typing import Any
+
+class EndpointHandler():
+    def __init__(self, path=""):
+        self.model = AutoModelForSeq2SeqLM.from_pretrained(f"{path}/4-bit", device_map="auto")
+        self.tokenizer = AutoTokenizer.from_pretrained(path)
+
+    def __call__(self, data: dict[str, Any]) -> dict[str, Any]:
+        inputs = data.get("inputs")
+        parameters = data.get("parameters")
+
+        if inputs is None:
+            raise ValueError(f"'inputs' is missing from the request body")
+
+        if not isinstance(inputs, str):
+            raise ValueError(f"Expected 'inputs' to be a str, but found {type(inputs)}")
+
+        if parameters is not None and not isinstance(parameters, dict):
+            raise ValueError(f"Expected 'parameters' to be a dict, but found {type(parameters)}")
+
+        # Truncate the tokens to 1024 to prevent errors with BART and long text.
+        tokens = self.tokenizer(
+            inputs,
+            max_length=1024,
+            truncation=True,
+            return_tensors="pt",
+            return_attention_mask=False,
+        )
+
+        # Ensure the input_ids and the model are both on the GPU to prevent errors.
+        input_ids = tokens.input_ids.to("cuda")
+
+        # Gradient calculation is not needed for inference.
+        with torch.no_grad():
+            if parameters is None:
+                output = self.model.generate(input_ids)
+            else:
+                output = self.model.generate(input_ids, **parameters)
+
+        generated_text = self.tokenizer.decode(output[0], skip_special_tokens=True)
+        return {"generated_text": generated_text}