Text Generation
llama-cpp-python
GGUF
English
code-generation
coding-assistant
llama.cpp
qwen2.5
python
javascript
fine-tuned
conversational
Instructions to use neuralbroker/blitzkode with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- llama-cpp-python
How to use neuralbroker/blitzkode with llama-cpp-python:
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="neuralbroker/blitzkode", filename="blitzkode.gguf", )
llm.create_chat_completion( messages = [ { "role": "user", "content": "What is the capital of France?" } ] ) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use neuralbroker/blitzkode with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf neuralbroker/blitzkode # Run inference directly in the terminal: llama cli -hf neuralbroker/blitzkode
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf neuralbroker/blitzkode # Run inference directly in the terminal: llama cli -hf neuralbroker/blitzkode
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf neuralbroker/blitzkode # Run inference directly in the terminal: ./llama-cli -hf neuralbroker/blitzkode
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf neuralbroker/blitzkode # Run inference directly in the terminal: ./build/bin/llama-cli -hf neuralbroker/blitzkode
Use Docker
docker model run hf.co/neuralbroker/blitzkode
- LM Studio
- Jan
- vLLM
How to use neuralbroker/blitzkode with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "neuralbroker/blitzkode" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "neuralbroker/blitzkode", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/neuralbroker/blitzkode
- Ollama
How to use neuralbroker/blitzkode with Ollama:
ollama run hf.co/neuralbroker/blitzkode
- Unsloth Studio
How to use neuralbroker/blitzkode with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for neuralbroker/blitzkode to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for neuralbroker/blitzkode to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for neuralbroker/blitzkode to start chatting
- Pi
How to use neuralbroker/blitzkode with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf neuralbroker/blitzkode
Configure the model in Pi
# Install Pi: npm install -g @mariozechner/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "neuralbroker/blitzkode" } ] } } }Run Pi
# Start Pi in your project directory: pi
- OpenClaw new
How to use neuralbroker/blitzkode with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf neuralbroker/blitzkode
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "neuralbroker/blitzkode" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
- Docker Model Runner
How to use neuralbroker/blitzkode with Docker Model Runner:
docker model run hf.co/neuralbroker/blitzkode
- Lemonade
How to use neuralbroker/blitzkode with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull neuralbroker/blitzkode
Run and chat with the model
lemonade run user.blitzkode-{{QUANT_TAG}}List all available models
lemonade list
- Hermes Agent
How to use neuralbroker/blitzkode with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf neuralbroker/blitzkode
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default neuralbroker/blitzkode
Run Hermes
hermes
- Atomic Chat
File size: 34,632 Bytes
b0bcd7a 6561a10 b0bcd7a 6561a10 6c79e9e b0bcd7a 6c79e9e 6561a10 6c79e9e b0bcd7a 6c79e9e 6561a10 b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6561a10 b0bcd7a 6c79e9e b0bcd7a 6c79e9e 6561a10 6c79e9e 6561a10 b0bcd7a 6561a10 b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6c79e9e b0bcd7a 6c79e9e 6561a10 b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6561a10 b0bcd7a 6561a10 b0bcd7a 6c79e9e b0bcd7a 6c79e9e 6561a10 b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6561a10 b0bcd7a 6561a10 6c79e9e 6561a10 b0bcd7a 6c79e9e b0bcd7a 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6561a10 6c79e9e b0bcd7a 6561a10 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6561a10 b0bcd7a 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e 6561a10 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e 6561a10 6c79e9e b0bcd7a 6c79e9e b0bcd7a 6c79e9e b0bcd7a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 | #!/usr/bin/env python3
"""
BlitzKode backend server.
Exposes a FastAPI backend for local GGUF inference through llama.cpp.
Model is loaded lazily so the module stays importable in tests and
environments where the model artifact is not present yet.
"""
from __future__ import annotations
import asyncio
import json
import logging
import os
import re
import threading
import time
import urllib.error
import urllib.parse
import urllib.request
from collections import deque
from collections.abc import Callable
from contextlib import asynccontextmanager, suppress
from dataclasses import dataclass
from dataclasses import field as dataclass_field
from html.parser import HTMLParser
from pathlib import Path
from typing import Any, Literal, cast
import llama_cpp
import uvicorn
from fastapi import FastAPI, Request
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse, StreamingResponse
from pydantic import BaseModel, Field
from starlette.middleware.base import BaseHTTPMiddleware
APP_NAME = "BlitzKode"
APP_VERSION = "2.0"
CREATOR = "Sajad"
ROOT_DIR = Path(__file__).resolve().parent
DEFAULT_MODEL_PATH = ROOT_DIR / "blitzkode.gguf"
DEFAULT_CONTEXT = 2048
DEFAULT_MAX_PROMPT_LENGTH = 4000
DEFAULT_MAX_TOKENS = 512
DEFAULT_RATE_LIMIT_MAX = 30
DEFAULT_MAX_SEARCH_RESULTS = 5
DEFAULT_SEARCH_TIMEOUT_SECONDS = 8
DEFAULT_SEARCH_CACHE_TTL_SECONDS = 300
DEFAULT_MAX_MESSAGES = 20
DEFAULT_BATCH = 256
DEFAULT_PROMPT_CACHE_BYTES = 64 * 1024 * 1024
STOP_TOKENS = ["<|im_end|>", "<|im_start|>user"]
SYSTEM_PROMPT = (
"<|im_start|>system\n"
"You are BlitzKode, an AI coding assistant created by Sajad. "
"You are an expert in Python, JavaScript, Java, C++, and other programming languages. "
"For coding work, first identify the user's goal, constraints, and any unknowns. "
"If asked about a library, API, file, function, citation, execution result, or repository detail that is not provided, "
"do not fabricate it. Say you do not know and explain how to verify it. "
"Prefer safe minimal fixes over speculative code. "
"Write clean, efficient, and well-documented code. Keep responses concise and practical.<|im_end|>"
)
logger = logging.getLogger("blitzkode")
def _bool_from_env(name: str, default: bool = False) -> bool:
value = os.getenv(name)
if value is None:
return default
return value.strip().lower() in {"1", "true", "yes", "on"}
def _int_from_env(name: str, default: int) -> int:
value = os.getenv(name)
if not value:
return default
try:
return int(value)
except ValueError:
return default
def _path_from_env(name: str, default: Path) -> Path:
value = os.getenv(name)
return Path(value) if value else default
def _validate_prompt(prompt: str, max_length: int) -> tuple[str, JSONResponse | None]:
prompt = prompt.strip()
if not prompt:
return prompt, JSONResponse({"error": "Prompt is required"}, status_code=400)
if len(prompt) > max_length:
return prompt, JSONResponse(
{"error": f"Prompt too long. Max {max_length} chars."},
status_code=400,
)
return prompt, None
_SIGNATURE_QUERY_RE = re.compile(
r"(?:signature|how\s+(?:do|can)\s+i\s+use|usage\s+of|docs?\s+for).{0,120}?"
r"(?P<symbol>[A-Za-z_][\w.]*\s*\([^)]*\)|[A-Za-z_][\w.]*\s+function)",
re.IGNORECASE | re.DOTALL,
)
_CODE_CONTEXT_RE = re.compile(r"```|\b(def|class|function|interface|type|import|from)\b|\{\s*\"|\bsource\s+code\b", re.IGNORECASE)
def _grounding_guard_response(prompt: str, has_external_context: bool = False) -> str | None:
"""Prevents confident fabricated API/function signatures when no source/docs are provided.
Small local models can ignore system instructions around unknown symbols. This guardrail only triggers for direct API/signature
lookup questions that contain no pasted source/docs context. It does not block normal code-generation tasks.
"""
if has_external_context or _CODE_CONTEXT_RE.search(prompt):
return None
match = _SIGNATURE_QUERY_RE.search(prompt)
if not match:
return None
symbol = " ".join(match.group("symbol").split())
return (
f"I don't have enough verified context to know the signature or usage of `{symbol}`. "
"Please provide the source code or official documentation, or enable research mode so I can ground the answer."
)
@dataclass(slots=True)
class Settings:
root_dir: Path = ROOT_DIR
model_path: Path = dataclass_field(default_factory=lambda: _path_from_env("BLITZKODE_MODEL_PATH", DEFAULT_MODEL_PATH))
host: str = os.getenv("BLITZKODE_HOST", "0.0.0.0")
port: int = _int_from_env("BLITZKODE_PORT", 7860)
n_gpu_layers: int = _int_from_env("BLITZKODE_GPU_LAYERS", 0)
n_ctx: int = _int_from_env("BLITZKODE_N_CTX", DEFAULT_CONTEXT)
n_threads: int = _int_from_env("BLITZKODE_THREADS", max(1, min(8, os.cpu_count() or 1)))
n_threads_batch: int = _int_from_env("BLITZKODE_THREADS_BATCH", max(1, min(8, os.cpu_count() or 1)))
n_batch: int = _int_from_env("BLITZKODE_BATCH", DEFAULT_BATCH)
n_ubatch: int = _int_from_env("BLITZKODE_UBATCH", min(DEFAULT_BATCH, 128))
prompt_cache_enabled: bool = _bool_from_env("BLITZKODE_PROMPT_CACHE", default=True)
prompt_cache_bytes: int = _int_from_env("BLITZKODE_PROMPT_CACHE_BYTES", DEFAULT_PROMPT_CACHE_BYTES)
use_mmap: bool = _bool_from_env("BLITZKODE_USE_MMAP", default=True)
use_mlock: bool = _bool_from_env("BLITZKODE_USE_MLOCK", default=False)
offload_kqv: bool = _bool_from_env("BLITZKODE_OFFLOAD_KQV", default=True)
max_prompt_length: int = _int_from_env("BLITZKODE_MAX_PROMPT_LENGTH", DEFAULT_MAX_PROMPT_LENGTH)
preload_model: bool = _bool_from_env("BLITZKODE_PRELOAD_MODEL", default=False)
cors_origins: str = os.getenv("BLITZKODE_CORS_ORIGINS", "http://localhost:7860")
api_key: str = os.getenv("BLITZKODE_API_KEY", "")
web_search_enabled: bool = _bool_from_env("BLITZKODE_WEB_SEARCH", default=True)
search_timeout_seconds: int = _int_from_env("BLITZKODE_SEARCH_TIMEOUT", DEFAULT_SEARCH_TIMEOUT_SECONDS)
max_search_results: int = _int_from_env("BLITZKODE_MAX_SEARCH_RESULTS", DEFAULT_MAX_SEARCH_RESULTS)
search_cache_ttl_seconds: int = _int_from_env("BLITZKODE_SEARCH_CACHE_TTL", DEFAULT_SEARCH_CACHE_TTL_SECONDS)
class MessageItem(BaseModel):
role: Literal["user", "assistant"]
content: str = Field(min_length=1, max_length=DEFAULT_MAX_PROMPT_LENGTH)
class GenerateRequest(BaseModel):
prompt: str
messages: list[MessageItem] = Field(default_factory=list, max_length=DEFAULT_MAX_MESSAGES)
temperature: float = Field(default=0.5, ge=0.0, le=2.0)
max_tokens: int = Field(default=256, ge=1, le=DEFAULT_MAX_TOKENS)
top_p: float = Field(default=0.95, gt=0.0, le=1.0)
top_k: int = Field(default=20, ge=1, le=200)
repeat_penalty: float = Field(default=1.05, ge=0.8, le=2.0)
class SearchRequest(BaseModel):
query: str = Field(min_length=1, max_length=500)
max_results: int = Field(default=DEFAULT_MAX_SEARCH_RESULTS, ge=1, le=10)
deep: bool = False
class ResearchGenerateRequest(GenerateRequest):
search_query: str | None = Field(default=None, max_length=500)
search_results: int = Field(default=DEFAULT_MAX_SEARCH_RESULTS, ge=1, le=10)
deep_search: bool = False
@dataclass(slots=True)
class SearchResult:
title: str
url: str
snippet: str
source: str = "DuckDuckGo"
def as_dict(self) -> dict[str, str]:
return {
"title": self.title,
"url": self.url,
"snippet": self.snippet,
"source": self.source,
}
class DuckDuckGoHTMLParser(HTMLParser):
def __init__(self, max_results: int):
super().__init__(convert_charrefs=True)
self.max_results = max_results
self.results: list[dict[str, str]] = []
self._active_field: Literal["title", "snippet"] | None = None
self._active_href = ""
self._text_parts: list[str] = []
def handle_starttag(self, tag: str, attrs: list[tuple[str, str | None]]) -> None:
if tag != "a":
return
attr_map = {name: value or "" for name, value in attrs}
classes = set(attr_map.get("class", "").split())
if "result__a" in classes:
if len(self.results) >= self.max_results:
return
self._active_field = "title"
elif "result__snippet" in classes:
self._active_field = "snippet"
else:
return
self._active_href = attr_map.get("href", "")
self._text_parts = []
def handle_data(self, data: str) -> None:
if self._active_field:
self._text_parts.append(data)
def handle_endtag(self, tag: str) -> None:
if tag != "a" or not self._active_field:
return
text = " ".join("".join(self._text_parts).split())
url = self._unwrap_result_url(self._active_href)
if self._active_field == "title" and text and url and len(self.results) < self.max_results:
self.results.append({"title": text, "url": url, "snippet": ""})
elif self._active_field == "snippet" and text and self.results:
target = next((item for item in reversed(self.results) if not item["snippet"]), None)
if target:
target["snippet"] = text
self._active_field = None
self._active_href = ""
self._text_parts = []
@staticmethod
def _unwrap_result_url(href: str) -> str:
href = href.strip()
if href.startswith("//"):
href = f"https:{href}"
parsed = urllib.parse.urlparse(href)
if "duckduckgo.com" in parsed.netloc and parsed.path.startswith("/l/"):
target = urllib.parse.parse_qs(parsed.query).get("uddg", [""])[0]
if target:
return target
return href
class WebSearchService:
def __init__(self, settings: Settings):
self.settings = settings
self._cache: dict[tuple[str, int, bool], tuple[float, list[dict[str, str]]]] = {}
self._cache_lock = threading.Lock()
@property
def enabled(self) -> bool:
return self.settings.web_search_enabled
def _query_variants(self, query: str, deep: bool) -> list[str]:
query = " ".join(query.split())
if not deep:
return [query]
return [
query,
f"{query} official documentation",
f"{query} best practices",
]
def _append_result(
self, results: list[SearchResult], seen_urls: set[str], title: str, url: str, snippet: str, max_results: int
) -> None:
title = " ".join((title or "Untitled").split())[:200]
url = (url or "").strip()
snippet = " ".join((snippet or "").split())[:500]
if not url or url in seen_urls or len(results) >= max_results:
return
seen_urls.add(url)
results.append(SearchResult(title=title, url=url, snippet=snippet))
def _collect_related_topics(self, topics: list[dict], results: list[SearchResult], seen_urls: set[str], max_results: int) -> None:
for topic in topics:
if len(results) >= max_results:
return
if "Topics" in topic:
self._collect_related_topics(topic.get("Topics", []), results, seen_urls, max_results)
continue
text = topic.get("Text", "")
url = topic.get("FirstURL", "")
if text and url:
title = text.split(" - ", 1)[0]
self._append_result(results, seen_urls, title, url, text, max_results)
def _read_search_payload(self, request: urllib.request.Request) -> dict[str, Any]:
with urllib.request.urlopen(request, timeout=self.settings.search_timeout_seconds) as response:
raw = response.read().decode("utf-8")
try:
payload = json.loads(raw)
except json.JSONDecodeError as exc:
raise RuntimeError("Search provider returned an invalid JSON response") from exc
if not isinstance(payload, dict):
raise RuntimeError("Search provider returned an unexpected response shape")
return payload
def _search_html(
self,
query: str,
results: list[SearchResult],
seen_urls: set[str],
max_results: int,
) -> None:
params = urllib.parse.urlencode({"q": query})
request = urllib.request.Request(
f"https://html.duckduckgo.com/html/?{params}",
headers={"User-Agent": f"Mozilla/5.0 {APP_NAME}/{APP_VERSION}"},
)
with urllib.request.urlopen(request, timeout=self.settings.search_timeout_seconds) as response:
raw = response.read().decode("utf-8", errors="replace")
parser = DuckDuckGoHTMLParser(max_results)
parser.feed(raw)
for item in parser.results:
self._append_result(results, seen_urls, item["title"], item["url"], item["snippet"], max_results)
def search(self, query: str, max_results: int = DEFAULT_MAX_SEARCH_RESULTS, deep: bool = False) -> list[dict[str, str]]:
if not self.enabled:
raise RuntimeError("Web search is disabled. Set BLITZKODE_WEB_SEARCH=true to enable it.")
query = " ".join(query.split())
if not query:
raise ValueError("Search query is required")
limit = min(max_results, max(1, self.settings.max_search_results), 10)
cache_key = (query.lower(), limit, deep)
now = time.monotonic()
with self._cache_lock:
cached = self._cache.get(cache_key)
if cached and now - cached[0] < self.settings.search_cache_ttl_seconds:
return [dict(item) for item in cached[1]]
results: list[SearchResult] = []
seen_urls: set[str] = set()
for variant in self._query_variants(query, deep):
if len(results) >= limit:
break
params = urllib.parse.urlencode(
{
"q": variant,
"format": "json",
"no_html": "1",
"skip_disambig": "1",
}
)
request = urllib.request.Request(
f"https://api.duckduckgo.com/?{params}",
headers={"User-Agent": f"{APP_NAME}/{APP_VERSION}"},
)
try:
payload = self._read_search_payload(request)
except RuntimeError as exc:
result_count = len(results)
self._search_html(variant, results, seen_urls, limit)
if len(results) == result_count:
raise exc
continue
self._append_result(
results,
seen_urls,
payload.get("Heading") or variant,
payload.get("AbstractURL", ""),
payload.get("AbstractText", ""),
limit,
)
self._collect_related_topics(payload.get("RelatedTopics", []), results, seen_urls, limit)
search_payload = [result.as_dict() for result in results]
with self._cache_lock:
self._cache[cache_key] = (time.monotonic(), search_payload)
return [dict(item) for item in search_payload]
class ModelService:
def __init__(self, settings: Settings):
self.settings = settings
self._llm: llama_cpp.Llama | None = None
self._init_lock = threading.Lock()
self._load_time_seconds: float | None = None
self._last_error: str | None = None
self._busy: bool = False
@property
def model_loaded(self) -> bool:
return self._llm is not None
@property
def model_exists(self) -> bool:
return self.settings.model_path.exists()
@property
def last_error(self) -> str | None:
return self._last_error
@property
def load_time_seconds(self) -> float | None:
return self._load_time_seconds
@property
def busy(self) -> bool:
return self._busy
def load_model(self):
if self._llm is not None:
return self._llm
with self._init_lock:
if self._llm is not None:
return self._llm
if not self.model_exists:
self._last_error = f"Model not found at {self.settings.model_path}"
raise FileNotFoundError(self._last_error)
start_time = time.perf_counter()
try:
self._llm = self._create_llama()
self._load_time_seconds = time.perf_counter() - start_time
self._last_error = None
self._configure_prompt_cache(self._llm)
logger.info(
"Model loaded in %.2fs (gpu_layers=%d, ctx=%d, threads=%d, batch=%d)",
self._load_time_seconds,
self.settings.n_gpu_layers,
self.settings.n_ctx,
self.settings.n_threads,
self.settings.n_batch,
)
except Exception as exc:
self._last_error = str(exc)
logger.error("Model load failed: %s", exc)
raise
return self._llm
def _create_llama(self) -> llama_cpp.Llama:
kwargs: dict[str, Any] = {
"model_path": str(self.settings.model_path),
"n_gpu_layers": self.settings.n_gpu_layers,
"n_ctx": self.settings.n_ctx,
"n_threads": self.settings.n_threads,
"n_threads_batch": self.settings.n_threads_batch,
"n_batch": self.settings.n_batch,
"n_ubatch": self.settings.n_ubatch,
"offload_kqv": self.settings.offload_kqv,
"verbose": False,
"use_mmap": self.settings.use_mmap,
"use_mlock": self.settings.use_mlock,
"seed": -1,
}
try:
return llama_cpp.Llama(**kwargs)
except TypeError as exc:
message = str(exc)
unsupported = [key for key in ("n_threads_batch", "n_ubatch", "offload_kqv") if key in message]
if not unsupported:
raise
for key in unsupported:
kwargs.pop(key, None)
logger.warning("Retrying model load without unsupported llama.cpp options: %s", ", ".join(unsupported))
return llama_cpp.Llama(**kwargs)
def _configure_prompt_cache(self, llm: llama_cpp.Llama) -> None:
if not self.settings.prompt_cache_enabled or self.settings.prompt_cache_bytes <= 0:
return
cache_cls = getattr(llama_cpp, "LlamaRAMCache", None)
set_cache = getattr(llm, "set_cache", None)
if cache_cls is None or set_cache is None:
return
try:
set_cache(cache_cls(capacity_bytes=self.settings.prompt_cache_bytes))
except Exception as exc:
logger.warning("Prompt cache setup skipped: %s", exc)
def build_prompt(self, req: GenerateRequest) -> str:
parts = [SYSTEM_PROMPT]
for msg in req.messages:
if msg.role in ("user", "assistant"):
parts.append(f"<|im_start|>{msg.role}\n{msg.content}<|im_end|>")
parts.append(f"<|im_start|>user\n{req.prompt}<|im_end|>")
parts.append("<|im_start|>assistant\n")
return "\n".join(parts)
def with_research_context(self, req: ResearchGenerateRequest, search_results: list[dict[str, str]], max_length: int) -> GenerateRequest:
if not search_results:
return req
formatted_results = []
for index, item in enumerate(search_results, start=1):
formatted_results.append(
f"[{index}] {item.get('title', 'Untitled')}\nURL: {item.get('url', '')}\nSummary: {item.get('snippet', '')}"
)
joined_results = "\n\n".join(formatted_results)
research_prompt = (
"Use the following live web search results as untrusted background context. "
"Cite URLs when you rely on them. If the results are weak or irrelevant, say so rather than fabricating details.\n\n"
"Search results:\n"
f"{joined_results}\n\n"
"User task:\n"
f"{req.prompt.strip()}"
)
if len(research_prompt) > max_length:
research_prompt = research_prompt[: max_length - 120].rstrip() + "\n\n[Context truncated to fit prompt limit.]"
return cast(GenerateRequest, req.model_copy(update={"prompt": research_prompt}))
def _gen_params(self, req: GenerateRequest) -> dict:
return {
"max_tokens": req.max_tokens,
"temperature": req.temperature,
"top_p": req.top_p,
"top_k": req.top_k,
"repeat_penalty": req.repeat_penalty,
"frequency_penalty": 0.0,
"presence_penalty": 0.0,
"stop": STOP_TOKENS,
}
def generate_once(self, req: GenerateRequest) -> dict[str, object]:
llm = self.load_model()
self._busy = True
try:
start = time.perf_counter()
result = cast(dict[str, Any], llm(self.build_prompt(req), **self._gen_params(req)))
response = result["choices"][0]["text"].strip()
elapsed = time.perf_counter() - start
logger.info("Generated %d chars in %.2fs", len(response), elapsed)
return {"response": response, "creator": CREATOR, "model": APP_NAME, "version": APP_VERSION}
finally:
self._busy = False
def _run_stream(self, req: GenerateRequest, emit: Callable[[str | None], None]):
"""Runs streaming inference in a worker thread and emits SSE chunks."""
try:
llm = self.load_model()
self._busy = True
start = time.perf_counter()
token_count = 0
stream = cast(Any, llm(self.build_prompt(req), stream=True, **self._gen_params(req)))
for token in stream:
if not token.get("choices"):
continue
text = token["choices"][0].get("text", "")
if text:
token_count += 1
emit(f"data: {json.dumps({'token': text})}\n\n")
elapsed = time.perf_counter() - start
logger.info("Streamed %d tokens in %.2fs", token_count, elapsed)
emit("data: [DONE]\n\n")
except Exception as exc:
logger.error("Stream error: %s", exc)
emit(f"data: {json.dumps({'error': str(exc)})}\n\n")
finally:
self._busy = False
emit(None)
def _check_api_key(request: Request, settings: Settings) -> JSONResponse | None:
if not settings.api_key:
return None
auth = request.headers.get("Authorization", "")
token = auth[7:] if auth.startswith("Bearer ") else auth
# Timing-safe comparison (prevent timing attacks)
import hmac
if not hmac.compare_digest(token, settings.api_key):
return JSONResponse({"error": "Unauthorized"}, status_code=401)
return None
class RateLimitMiddleware(BaseHTTPMiddleware):
def __init__(self, app, max_requests: int = DEFAULT_RATE_LIMIT_MAX, window_seconds: int = 60):
super().__init__(app)
self._max = max_requests
self._window = window_seconds
self._clients: dict[str, deque[float]] = {}
self._lock = threading.Lock()
self._cleanup_done = 0
async def dispatch(self, request: Request, call_next):
client_ip = request.client.host if request.client else "unknown"
now = time.monotonic()
# Cleanup old entries periodically (every 1000 requests)
self._cleanup_done += 1
if self._cleanup_done > 1000:
self._cleanup_done = 0
with self._lock:
cutoff = now - self._window
self._clients = {ip: deque(t for t in ts if t >= cutoff) for ip, ts in self._clients.items() if ts}
with self._lock:
timestamps = self._clients.setdefault(client_ip, deque())
cutoff = now - self._window
while timestamps and timestamps[0] < cutoff:
timestamps.popleft()
if len(timestamps) >= self._max:
return JSONResponse(
{"error": "Rate limit exceeded. Try again later."},
status_code=429,
headers={"Retry-After": str(self._window)},
)
timestamps.append(now)
return await call_next(request)
class RequestSizeLimitMiddleware(BaseHTTPMiddleware):
def __init__(self, app, max_bytes: int = 50_000):
super().__init__(app)
self._max = max_bytes
async def dispatch(self, request: Request, call_next):
content_length = request.headers.get("content-length")
if content_length:
try:
if int(content_length) > self._max:
return JSONResponse({"error": "Request body too large"}, status_code=413)
except ValueError:
return JSONResponse({"error": "Invalid Content-Length header"}, status_code=400)
return await call_next(request)
def create_app(settings: Settings | None = None) -> FastAPI:
settings = settings or Settings()
model_service = ModelService(settings)
search_service = WebSearchService(settings)
model_lock = asyncio.Lock()
logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(name)s] %(message)s", datefmt="%H:%M:%S")
@asynccontextmanager
async def lifespan(_: FastAPI):
if settings.preload_model:
with suppress(Exception):
await asyncio.to_thread(model_service.load_model)
yield
app = FastAPI(title=f"{APP_NAME} API", version=APP_VERSION, lifespan=lifespan)
app.state.settings = settings
app.state.model_service = model_service
app.state.search_service = search_service
cors_origins = [o.strip() for o in settings.cors_origins.split(",") if o.strip()]
app.add_middleware(
CORSMiddleware,
allow_origins=cors_origins,
allow_methods=["POST", "GET", "OPTIONS"],
allow_headers=["Content-Type", "Authorization"],
)
if _bool_from_env("BLITZKODE_RATE_LIMIT", default=True):
app.add_middleware(RateLimitMiddleware, max_requests=_int_from_env("BLITZKODE_RATE_LIMIT_MAX", DEFAULT_RATE_LIMIT_MAX))
app.add_middleware(RequestSizeLimitMiddleware, max_bytes=_int_from_env("BLITZKODE_MAX_REQUEST_BYTES", 50_000))
@app.get("/")
async def root():
return JSONResponse(
{
"name": APP_NAME,
"version": APP_VERSION,
"message": "BlitzKode backend API is running. Use /info for endpoint details.",
}
)
@app.get("/health")
async def health():
status = "healthy" if model_service.model_exists and not model_service.last_error else "degraded"
return JSONResponse(
{
"status": status,
"model_loaded": model_service.model_loaded,
"model_path": str(settings.model_path),
"model_exists": model_service.model_exists,
"version": APP_VERSION,
"gpu_layers": settings.n_gpu_layers,
"last_error": model_service.last_error,
"busy": model_service.busy,
}
)
@app.post("/generate")
async def generate(req: GenerateRequest, request: Request):
auth_err = _check_api_key(request, settings)
if auth_err:
return auth_err
prompt, err = _validate_prompt(req.prompt, settings.max_prompt_length)
if err:
return err
guard_response = _grounding_guard_response(prompt)
if guard_response:
return JSONResponse(
{"response": guard_response, "creator": CREATOR, "model": APP_NAME, "version": APP_VERSION, "guarded": True}
)
async with model_lock:
try:
sanitized = req.model_copy(update={"prompt": prompt})
payload = await asyncio.to_thread(model_service.generate_once, sanitized)
return JSONResponse(payload)
except FileNotFoundError as exc:
return JSONResponse({"error": str(exc)}, status_code=503)
except Exception as exc:
return JSONResponse({"error": str(exc)}, status_code=500)
@app.post("/generate/research")
async def generate_research(req: ResearchGenerateRequest, request: Request):
auth_err = _check_api_key(request, settings)
if auth_err:
return auth_err
prompt, err = _validate_prompt(req.prompt, settings.max_prompt_length)
if err:
return err
if not search_service.enabled:
return JSONResponse({"error": "Web search is disabled"}, status_code=503)
search_query = (req.search_query or prompt).strip()
try:
results = await asyncio.to_thread(search_service.search, search_query, req.search_results, req.deep_search)
sanitized = req.model_copy(update={"prompt": prompt})
enriched = model_service.with_research_context(sanitized, results, settings.max_prompt_length)
async with model_lock:
payload = await asyncio.to_thread(model_service.generate_once, enriched)
payload["search_results"] = results
return JSONResponse(payload)
except FileNotFoundError as exc:
return JSONResponse({"error": str(exc)}, status_code=503)
except (RuntimeError, urllib.error.URLError, TimeoutError) as exc:
return JSONResponse({"error": f"Search failed: {exc}"}, status_code=502)
except Exception as exc:
return JSONResponse({"error": str(exc)}, status_code=500)
@app.post("/generate/stream")
async def generate_stream(req: GenerateRequest, request: Request):
auth_err = _check_api_key(request, settings)
if auth_err:
return auth_err
prompt, err = _validate_prompt(req.prompt, settings.max_prompt_length)
if err:
return err
if not model_service.model_exists:
return JSONResponse({"error": f"Model not found at {settings.model_path}"}, status_code=503)
guard_response = _grounding_guard_response(prompt)
if guard_response:
async def _guarded_stream():
yield f"data: {json.dumps({'token': guard_response})}\n\n"
yield "data: [DONE]\n\n"
return StreamingResponse(_guarded_stream(), media_type="text/event-stream")
sanitized = req.model_copy(update={"prompt": prompt})
async def _locked_stream():
async with model_lock:
loop = asyncio.get_running_loop()
token_q: asyncio.Queue[str | None] = asyncio.Queue()
def emit(chunk: str | None) -> None:
loop.call_soon_threadsafe(token_q.put_nowait, chunk)
thread = threading.Thread(
target=model_service._run_stream,
args=(sanitized, emit),
daemon=True,
)
thread.start()
while True:
chunk = await token_q.get()
if chunk is None:
break
yield chunk
return StreamingResponse(
_locked_stream(),
media_type="text/event-stream",
headers={"Cache-Control": "no-cache", "Connection": "keep-alive", "X-Accel-Buffering": "no"},
)
@app.post("/search/web")
async def search_web(req: SearchRequest, request: Request):
auth_err = _check_api_key(request, settings)
if auth_err:
return auth_err
if not search_service.enabled:
return JSONResponse({"error": "Web search is disabled"}, status_code=503)
try:
results = await asyncio.to_thread(search_service.search, req.query, req.max_results, req.deep)
return JSONResponse({"query": req.query.strip(), "deep": req.deep, "results": results})
except (RuntimeError, urllib.error.URLError, TimeoutError) as exc:
return JSONResponse({"error": f"Search failed: {exc}"}, status_code=502)
except Exception as exc:
return JSONResponse({"error": str(exc)}, status_code=500)
@app.get("/info")
async def info():
return JSONResponse(
{
"name": APP_NAME,
"creator": CREATOR,
"version": APP_VERSION,
"status": "ready" if model_service.model_exists else "model-missing",
"mode": f"{'GPU' if settings.n_gpu_layers > 0 else 'CPU'} (llama.cpp)",
"gpu_layers": settings.n_gpu_layers,
"context_window": settings.n_ctx,
"threads": settings.n_threads,
"threads_batch": settings.n_threads_batch,
"batch": settings.n_batch,
"ubatch": settings.n_ubatch,
"prompt_cache_enabled": settings.prompt_cache_enabled,
"model_loaded": model_service.model_loaded,
"load_time_seconds": model_service.load_time_seconds,
"busy": model_service.busy,
"web_search_enabled": search_service.enabled,
"endpoints": {
"generate": "POST /generate",
"research_generate": "POST /generate/research",
"stream": "POST /generate/stream",
"search": "POST /search/web",
"health": "GET /health",
"info": "GET /info",
},
}
)
return app
app = create_app()
def main() -> None:
s = Settings()
print(f"\n{'=' * 50}")
print(f"{APP_NAME.upper()} v{APP_VERSION}")
print(f"Creator: {CREATOR}")
print(f"{'=' * 50}")
print(f"Model: {s.model_path}")
print(f"GPU: {s.n_gpu_layers} layers")
print(f"Ctx: {s.n_ctx} | Threads: {s.n_threads}")
print(f"URL: http://localhost:{s.port}\n")
uvicorn.run(app, host=s.host, port=s.port, log_level="warning")
if __name__ == "__main__":
main()
|