Upload folder using huggingface_hub

README.md

@@ -0,0 +1,68 @@
+---
+language:
+- en
+tags:
+- causal-lm
+- llama
+- instruction-tuned
+- point-in-time
+- dated
+- lookahead-bias-free
+pipeline_tag: text-generation
+---
+
+# DatedGPT-2019-Instruct
+
+**DatedGPT** is a family of point-in-time language models: each vintage is
+trained only on data available up to its cutoff date, making it suitable for
+lookahead-bias-free prediction and point-in-time analysis.
+
+This is the **instruction-tuned chat model** with data up to **2019**.
+For the base (pretrained) model, see
+[datedgpt/datedgpt-2019-base](https://huggingface.co/datedgpt/datedgpt-2019-base).
+
+| Property | Value |
+|----------|-------|
+| Architecture | LlamaForCausalLM |
+| Parameters | ~1.3 B |
+| Context length | 2048 |
+| Vocab | 32,000 (SentencePiece) |
+| Precision | bfloat16 |
+| Data vintage | 2019 |
+
+## Chat template
+
+The Llama-2-style chat template ships in `tokenizer_config.json` — apply it
+with the tokenizer. The BOS token must come from the tokenizer, **not** as a
+literal `"<s>"` string in your prompt text.
+
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+repo_id = "datedgpt/datedgpt-2019-instruct"
+tokenizer = AutoTokenizer.from_pretrained(repo_id)
+model = AutoModelForCausalLM.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map="auto")
+
+prompt = tokenizer.apply_chat_template(
+    [{"role": "user", "content": "What is the capital of France?"}],
+    tokenize=False,
+)
+inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+output = model.generate(**inputs, max_new_tokens=128, do_sample=True,
+                        temperature=0.7, top_p=0.95, use_cache=True,
+                        eos_token_id=tokenizer.eos_token_id,
+                        pad_token_id=tokenizer.eos_token_id)
+print(tokenizer.decode(output[0, inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
+```
+
+## Serving on HF Inference Endpoints
+
+This repo ships a `handler.py` that applies the chat template server-side —
+clients send plain text (or a messages list for multi-turn). Deploy with the
+Default container on a bf16-capable GPU (A10G or better).
+
+## Limitations
+
+- Knowledge limited to the 2019 data vintage.
+- No RLHF or safety tuning; outputs can be confidently wrong.

config.json

@@ -0,0 +1,29 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5504,
+  "max_position_embeddings": 2048,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 16,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.0",
+  "use_cache": false,
+  "vocab_size": 32000
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "transformers_version": "4.51.0"
+}

handler.py

@@ -0,0 +1,99 @@
+import os
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+# Keep only the last round (1 user+assistant pair) + the current user message
+MAX_HISTORY_MESSAGES = 3
+
+
+class EndpointHandler:
+    """
+    Hugging Face Inference Endpoints custom handler.
+
+    Expects input like:
+      - {"inputs": "hello"}                                  -> single-turn, auto-wrapped with chat template
+      - {"inputs": [{"role":"user","content":"hello"}, ...]}  -> multi-turn chat (messages list)
+      - {"inputs": "hello", "parameters": {"raw": true}}     -> sent as-is (no template)
+
+    Optional:
+      - {"parameters": {"max_new_tokens": 512, "temperature": 0.7, ...}}
+    """
+
+    def __init__(self, path: str = ""):
+        model_dir = path or os.getenv("HF_MODEL_DIR", ".")
+        self.device = "cuda" if torch.cuda.is_available() else "cpu"
+
+        self.tokenizer = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True)
+
+        # Ensure pad token exists (common for causal LMs)
+        if self.tokenizer.pad_token is None:
+            self.tokenizer.pad_token = self.tokenizer.eos_token
+            self.tokenizer.pad_token_id = self.tokenizer.eos_token_id
+
+
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_dir,
+            torch_dtype="auto",
+            device_map="auto" if torch.cuda.is_available() else None,
+            trust_remote_code=True,
+        )
+        self.model.eval()
+
+    def __call__(self, data: dict) -> dict:
+        inputs = data.get("inputs", data)
+        params = data.get("parameters", {}) or {}
+
+        raw = bool(params.pop("raw", False))
+
+        if raw:
+            if not isinstance(inputs, str):
+                raise ValueError("raw mode requires inputs to be a string.")
+            prompt = inputs
+        elif isinstance(inputs, list):
+            # Multi-turn: inputs is a list of {"role": ..., "content": ...}
+            inputs = inputs[-MAX_HISTORY_MESSAGES:]
+            prompt = self.tokenizer.apply_chat_template(
+                inputs, tokenize=False,
+            )
+        elif isinstance(inputs, str):
+            # Single-turn: wrap in a one-message list
+            prompt = self.tokenizer.apply_chat_template(
+                [{"role": "user", "content": inputs}],
+                tokenize=False,
+            )
+        else:
+            raise ValueError("inputs must be a string or a list of messages.")
+
+        enc = self.tokenizer(
+            prompt,
+            return_tensors="pt",
+            padding=False,
+            truncation=True,
+        )
+        input_ids = enc["input_ids"].to(self.model.device)
+        attention_mask = enc.get("attention_mask", torch.ones_like(input_ids)).to(self.model.device)
+
+        gen_kwargs = {
+            "max_new_tokens": min(int(params.pop("max_new_tokens", 512)), 512),
+            "do_sample": bool(params.pop("do_sample", True)),
+            "temperature": float(params.pop("temperature", 0.7)),
+            "top_p": float(params.pop("top_p", 0.95)),
+            "repetition_penalty": float(params.pop("repetition_penalty", 1.2)),
+            "eos_token_id": self.tokenizer.eos_token_id,
+            "pad_token_id": self.tokenizer.pad_token_id,
+        }
+        gen_kwargs.update(params)
+
+        with torch.no_grad():
+            out = self.model.generate(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                **gen_kwargs,
+            )
+
+        # Return only newly generated tokens (your current behavior)
+        new_tokens = out[0, input_ids.shape[-1]:]
+        text = self.tokenizer.decode(new_tokens, skip_special_tokens=True)
+
+        return {"generated_text": text}
+

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f4a342eb0cf7f7060794554f83f2ab744a41c629d018dd74f9f784b363c2894
+size 2690871976

requirements.txt

@@ -0,0 +1,2 @@
+accelerate
+safetensors

special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,45 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "chat_template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% endif %}{% for message in loop_messages %}{% if loop.index0 == 0 and system_message is defined %}{% set content = '<<SYS>>\n' + system_message + '\n<</SYS>>\n\n' + message['content'] %}{% else %}{% set content = message['content'] %}{% endif %}{% if message['role'] == 'user' %}{{ '<s>' + '[INST] ' + content + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ content + '</s>' }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 2048,
+  "pad_token": "</s>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "split_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}