inference.py

#!/usr/bin/env python3
"""Standalone chat inference for kerzgrr/Tercet-R-1.1.

Streams a live reasoning trace (`<think>` / `<|think|>`) and runs SmolTalk
tool-call rounds. Built-in `web-search`, `python`, and `calculator` tools
execute automatically. Other observations are sent as `<|tool_response|>`.

Examples:
  python inference.py --prompt "What is the capital of France?"
  python inference.py --tools web-search,python,calculator
  python inference.py --no-think --prompt "Reply in one sentence."
"""

from __future__ import annotations

import argparse
import json
import os
import platform
import shutil
import subprocess
import sys
import time
import warnings
from dataclasses import fields
from pathlib import Path
from typing import Any

import torch
from safetensors.torch import load_file
from tokenizers import Tokenizer


def _silence_runtime_warnings() -> None:
    patterns = (
        r"tl\.make_block_ptr is deprecated",
        r"Memory efficient kernel not used because",
        r"Memory Efficient attention has been runtime disabled",
        r"Flash attention kernel not used because",
        r"Torch was not compiled with flash attention",
        r"cuDNN attention kernel not used because",
        r"cuDNN attention has been runtime disabled",
    )
    for pattern in patterns:
        warnings.filterwarnings("ignore", message=pattern)


REPO_ID = "kerzgrr/Tercet-R-1.1"
FLA_COMMIT = "cbb0a72efb55c18ca0ef4f298298317573ad2cb3"
FLA_REPO = "https://github.com/fla-org/flash-linear-attention.git"
PATCH_FILES = (
    "fla/__init__.py",
    "fla/ops/__init__.py",
    "fla/layers/__init__.py",
    "fla/ops/simple_gla/__init__.py",
)
TINY_GDN_FILES = (
    "tiny_gdn/__init__.py",
    "tiny_gdn/config.py",
    "tiny_gdn/model.py",
    "tiny_gdn/tools.py",
    "tiny_gdn/smoltalk_chat.py",
    "tiny_gdn/chatml.py",
    "tiny_gdn/cli_chat.py",
    "tiny_gdn/detokenize.py",
    "tiny_gdn/code_exec.py",
    "tiny_gdn/web_search.py",
)


def _download(filename: str, local_dir: Path | None) -> Path:
    from huggingface_hub import hf_hub_download

    return Path(
        hf_hub_download(
            repo_id=REPO_ID,
            filename=filename,
            local_dir=str(local_dir) if local_dir else None,
        )
    )


def _run(cmd: list[str], *, cwd: Path | None = None, env: dict | None = None) -> None:
    print("+", " ".join(cmd), flush=True)
    merged = os.environ.copy()
    if env:
        merged.update(env)
    merged.setdefault("PYTHONUTF8", "1")
    merged.setdefault("PYTHONIOENCODING", "utf-8")
    subprocess.check_call(cmd, cwd=str(cwd) if cwd else None, env=merged)


def _pip_install(*args: str) -> None:
    _run([sys.executable, "-m", "pip", "install", *args])


def _fla_importable() -> tuple[bool, str]:
    try:
        from fla.layers.gdn2 import GatedDeltaNet2  # noqa: F401
    except Exception as error:  # noqa: BLE001
        return False, str(error)
    return True, ""


def _cache_root() -> Path:
    override = os.environ.get("MONOSTICH_CACHE")
    if override:
        path = Path(override).expanduser().resolve()
    else:
        path = Path.home() / ".cache" / "tercet-r"
    path.mkdir(parents=True, exist_ok=True)
    return path


def _ensure_git() -> None:
    if shutil.which("git") is None:
        raise RuntimeError(
            "git is required to auto-install flash-linear-attention. "
            "Install Git and ensure it is on PATH."
        )


def _apply_windows_fla_patches(fla_root: Path, local_dir: Path | None) -> None:
    print("Applying Windows FLA import patches from the Hub …", flush=True)
    for relative in PATCH_FILES:
        source = _download(f"windows_fla_patches/{relative}", local_dir)
        target = fla_root / relative
        target.parent.mkdir(parents=True, exist_ok=True)
        shutil.copy2(source, target)
        print(f"  patched {relative}", flush=True)


def _install_fla(local_dir: Path | None) -> None:
    print("flash-linear-attention missing/broken — installing automatically …", flush=True)
    _pip_install("einops", "numpy")
    if platform.system() != "Windows":
        _pip_install("--no-deps", f"git+{FLA_REPO}@{FLA_COMMIT}")
        return

    _ensure_git()
    fla_root = _cache_root() / "flash-linear-attention"
    if (fla_root / ".git").is_dir():
        _run(["git", "fetch", "--depth", "1", "origin", FLA_COMMIT], cwd=fla_root)
        _run(["git", "checkout", "--force", FLA_COMMIT], cwd=fla_root)
    else:
        if fla_root.exists():
            shutil.rmtree(fla_root)
        _run(["git", "clone", "--filter=blob:none", FLA_REPO, str(fla_root)])
        _run(["git", "fetch", "--depth", "1", "origin", FLA_COMMIT], cwd=fla_root)
        _run(["git", "checkout", "--force", FLA_COMMIT], cwd=fla_root)

    _apply_windows_fla_patches(fla_root, local_dir)
    _pip_install("--no-build-isolation", "--no-deps", "-e", str(fla_root))


def _ensure_fla(local_dir: Path | None) -> None:
    ok, error = _fla_importable()
    if ok:
        return
    print(f"FLA not ready ({error})", flush=True)
    try:
        _install_fla(local_dir)
    except Exception as install_error:  # noqa: BLE001
        raise RuntimeError(
            "Automatic flash-linear-attention install failed.\n"
            f"Original import error: {error}\n"
            f"Install error: {install_error}"
        ) from install_error

    for name in list(sys.modules):
        if name == "fla" or name.startswith("fla."):
            del sys.modules[name]

    ok, error = _fla_importable()
    if not ok:
        raise RuntimeError(
            "flash-linear-attention installed but still failed to import "
            f"GatedDeltaNet2: {error}"
        )
    print("flash-linear-attention ready.", flush=True)


def _ensure_tiny_gdn(local_dir: Path | None) -> Path:
    here = Path(__file__).resolve().parent
    if (here / "tiny_gdn" / "cli_chat.py").is_file():
        return here
    if local_dir and (local_dir / "tiny_gdn" / "cli_chat.py").is_file():
        return local_dir
    for name in TINY_GDN_FILES:
        _download(name, local_dir)
    return _download("tiny_gdn/__init__.py", local_dir).parent.parent


def _sample(
    logits: torch.Tensor,
    *,
    temperature: float,
    top_p: float,
    top_k: int,
    generator: torch.Generator,
) -> int:
    logits = logits.float()
    if temperature <= 1e-5:
        return int(torch.argmax(logits).item())
    logits = logits / temperature
    if 0 < top_k < logits.shape[-1]:
        threshold = torch.topk(logits, top_k).values[-1]
        logits = logits.masked_fill(logits < threshold, -torch.inf)
    if top_p < 1.0:
        sorted_logits, sorted_indices = torch.sort(logits, descending=True)
        probs = torch.softmax(sorted_logits, dim=-1)
        remove = torch.cumsum(probs, dim=-1) > top_p
        remove[1:] = remove[:-1].clone()
        remove[0] = False
        sorted_logits = sorted_logits.masked_fill(remove, -torch.inf)
        logits = torch.full_like(logits, -torch.inf)
        logits.scatter_(0, sorted_indices, sorted_logits)
    probs = torch.softmax(logits, dim=-1).cpu()
    return int(torch.multinomial(probs, 1, generator=generator).item())


def _apply_repetition_penalty(
    logits: torch.Tensor,
    token_ids: list[int],
    penalty: float,
    window: int,
) -> torch.Tensor:
    if penalty == 1.0 or not token_ids:
        return logits
    recent = token_ids[-window:] if window > 0 else token_ids
    unique = torch.tensor(list(set(recent)), dtype=torch.long, device=logits.device)
    score = logits[unique]
    logits[unique] = torch.where(score > 0, score / penalty, score * penalty)
    return logits


@torch.inference_mode()
def generate_turn(
    model,
    tokenizer: Tokenizer,
    encode_chatml_messages,
    IncrementalUtf8Decoder,
    parse_tool_calls,
    streamer,
    messages: list[dict[str, Any]],
    *,
    tools: Any | None,
    enable_thinking: bool,
    max_new_tokens: int,
    context_length: int,
    temperature: float,
    top_p: float,
    top_k: int,
    repetition_penalty: float,
    repetition_window: int,
    seed: int,
    device: torch.device,
) -> tuple[str, tuple]:
    prompt_ids = encode_chatml_messages(
        tokenizer,
        messages,
        tools=tools,
        enable_thinking=enable_thinking,
    )
    eos_id = int(model.config.eos_token_id)
    im_end = tokenizer.token_to_id("<|im_end|>")
    stop = {eos_id}
    if im_end is not None:
        stop.add(im_end)

    token_ids = list(prompt_ids[-context_length:])
    generated: list[int] = []
    decoder = IncrementalUtf8Decoder(tokenizer)
    stop_reason = "max_new_tokens"
    generator = torch.Generator(device="cpu")
    generator.manual_seed(seed)
    started = time.perf_counter()

    input_ids = torch.tensor([token_ids], dtype=torch.long, device=device)
    output = model(
        input_ids,
        return_logits=True,
        logits_to_keep=1,
        use_cache=True,
    )

    for step in range(max_new_tokens):
        if output.logits is None:
            raise RuntimeError("Model returned no logits")
        next_logits = _apply_repetition_penalty(
            output.logits[0, -1],
            token_ids,
            repetition_penalty,
            repetition_window,
        )
        next_id = _sample(
            next_logits,
            temperature=temperature,
            top_p=top_p,
            top_k=top_k,
            generator=generator,
        )
        if next_id in stop:
            stop_reason = "stop"
            break
        token_ids.append(next_id)
        generated.append(next_id)
        decoder.push(next_id)
        if streamer is not None:
            streamer.update(decoder.text)
        if step + 1 < max_new_tokens:
            if output.past_key_values is None:
                raise RuntimeError("Model returned no decode cache")
            input_ids = torch.tensor([[next_id]], dtype=torch.long, device=device)
            output = model(
                input_ids,
                return_logits=True,
                logits_to_keep=1,
                past_key_values=output.past_key_values,
                use_cache=True,
            )

    decoded = decoder.finalize()
    if streamer is not None:
        streamer.finish()
    elapsed = time.perf_counter() - started
    tps = len(generated) / max(elapsed, 1e-9)
    print(
        f"[done] tokens={len(generated)} stop={stop_reason} {tps:.1f} tok/s",
        file=sys.stderr,
        flush=True,
    )
    return decoded, tuple(parse_tool_calls(decoded))


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description="Tercet-R-1.1 chat inference")
    parser.add_argument("--prompt", default=None, help="Single user prompt")
    parser.add_argument("--system", default="", help="Optional system prompt")
    parser.add_argument(
        "--think",
        dest="enable_thinking",
        action="store_true",
        default=True,
    )
    parser.add_argument(
        "--no-think",
        dest="enable_thinking",
        action="store_false",
    )
    parser.add_argument(
        "--tools",
        default="",
        help="Built-in tools: web-search, python, calculator (comma-separated)",
    )
    parser.add_argument("--tools-json", default=None)
    parser.add_argument("--max-tool-rounds", type=int, default=8)
    parser.add_argument("--max-new-tokens", type=int, default=4096)
    parser.add_argument("--temperature", type=float, default=0.7)
    parser.add_argument("--top-p", type=float, default=0.9)
    parser.add_argument("--top-k", type=int, default=50)
    parser.add_argument("--repetition-penalty", type=float, default=1.08)
    parser.add_argument("--repetition-window", type=int, default=256)
    parser.add_argument("--context-length", type=int, default=16384)
    parser.add_argument("--seed", type=int, default=42)
    parser.add_argument(
        "--device",
        default="cuda" if torch.cuda.is_available() else "cpu",
        choices=["cuda", "cpu"],
    )
    parser.add_argument("--no-stream", action="store_true")
    parser.add_argument("--no-color", action="store_true")
    parser.add_argument("--local-dir", default=None)
    parser.add_argument("--repo-id", default=REPO_ID)
    return parser.parse_args()


def main() -> int:
    _silence_runtime_warnings()
    args = parse_args()
    global REPO_ID
    REPO_ID = args.repo_id
    local_dir = Path(args.local_dir).resolve() if args.local_dir else None

    print(f"Loading Tercet-R-1.1 from huggingface.co/{REPO_ID} …", flush=True)
    try:
        package_root = _ensure_tiny_gdn(local_dir)
    except Exception as error:  # noqa: BLE001
        print(f"Failed to resolve tiny_gdn package: {error}", file=sys.stderr)
        return 1

    if str(package_root) not in sys.path:
        sys.path.insert(0, str(package_root))

    try:
        _ensure_fla(local_dir)
    except Exception as error:  # noqa: BLE001
        print(str(error), file=sys.stderr)
        return 1

    try:
        from tiny_gdn import TinyGDNConfig, TinyGDNForCausalLM
        from tiny_gdn.chatml import encode_chatml_messages
        from tiny_gdn.cli_chat import (
            ChatLoopState,
            LiveReasoningStreamer,
            append_tool_round,
            apply_slash_command,
            prompt_tool_results,
            resolve_cli_tools,
        )
        from tiny_gdn.detokenize import IncrementalUtf8Decoder
        from tiny_gdn.tools import parse_tool_calls
    except ImportError as error:
        print(f"Could not import tiny_gdn chat stack: {error}", file=sys.stderr)
        return 1

    try:
        raw_json = (
            Path(args.tools_json).read_text(encoding="utf-8")
            if args.tools_json
            else None
        )
        tools = resolve_cli_tools(args.tools, raw_json)
    except (OSError, ValueError, json.JSONDecodeError) as error:
        print(error, file=sys.stderr)
        return 1

    weights_path = _download("model.safetensors", local_dir)
    tok_path = _download("tokenizer.json", local_dir)
    cfg_path = _download("config.json", local_dir)

    raw = json.loads(cfg_path.read_text(encoding="utf-8"))
    allowed = {item.name for item in fields(TinyGDNConfig)}
    payload = {key: value for key, value in raw.items() if key in allowed}
    if "shared_layer_indices" in payload:
        payload["shared_layer_indices"] = tuple(payload["shared_layer_indices"])
    config = TinyGDNConfig(**payload)

    device = torch.device(args.device)
    if device.type == "cuda" and not torch.cuda.is_available():
        print("CUDA requested but unavailable; falling back to CPU.", flush=True)
        device = torch.device("cpu")
    dtype = torch.bfloat16 if device.type == "cuda" else torch.float32

    print(
        f"Building TinyGDN ({config.num_hidden_layers}L / {config.hidden_size}d) "
        f"on {device} …",
        flush=True,
    )
    model = TinyGDNForCausalLM(config)
    state = load_file(str(weights_path), device="cpu")
    model.load_state_dict(state, strict=True)
    del state
    model = model.to(device=device, dtype=dtype)
    model.eval()
    model.requires_grad_(False)

    tokenizer = Tokenizer.from_file(str(tok_path))
    context_length = min(args.context_length, config.max_position_embeddings)
    color = not args.no_color and sys.stdout.isatty()

    def read_line(prompt: str) -> str:
        return input(prompt)

    def run_turn(state: ChatLoopState) -> None:
        for round_index in range(max(1, args.max_tool_rounds + 1)):
            streamer = (
                None
                if args.no_stream
                else LiveReasoningStreamer(sys.stdout, color=color)
            )
            if streamer is not None:
                print("assistant> ", end="", flush=True)
            text, calls = generate_turn(
                model,
                tokenizer,
                encode_chatml_messages,
                IncrementalUtf8Decoder,
                parse_tool_calls,
                streamer,
                state.messages,
                tools=state.tools,
                enable_thinking=state.enable_thinking,
                max_new_tokens=args.max_new_tokens,
                context_length=context_length,
                temperature=args.temperature,
                top_p=args.top_p,
                top_k=args.top_k,
                repetition_penalty=args.repetition_penalty,
                repetition_window=args.repetition_window,
                seed=args.seed,
                device=device,
            )
            if args.no_stream:
                print(f"assistant> {text}")
            if not calls or round_index >= args.max_tool_rounds:
                state.messages.append({"role": "assistant", "content": text})
                return
            results = prompt_tool_results(calls, read_line=read_line)
            append_tool_round(state.messages, text, results)

    state = ChatLoopState(
        system=args.system,
        enable_thinking=args.enable_thinking,
        tools=tools,
    )
    state.reset()

    if args.prompt is not None:
        state.add_user(args.prompt)
        run_turn(state)
        return 0

    print(
        "Interactive Tercet-R-1.1 chat. "
        "Commands: /think  /no_think  /system …  /reset  /exit\n"
        "web-search, python, and calculator run automatically when enabled. "
        "Other <tool_call> observations can be pasted at the prompt.",
        flush=True,
    )
    while True:
        try:
            user_input = input("user> ")
        except (EOFError, KeyboardInterrupt):
            print()
            return 0
        text = user_input.strip()
        if not text:
            continue
        status = apply_slash_command(state, text)
        if status == "exit":
            return 0
        if status is not None:
            print(f"({status})", flush=True)
            continue
        state.add_user(text)
        run_turn(state)
        print(flush=True)
    return 0


if __name__ == "__main__":
    raise SystemExit(main())