File size: 13,302 Bytes
eb5a3fb
 
 
47031d7
eb5a3fb
47031d7
63fdbaa
a4e24d4
db5664e
eb5a3fb
db5664e
d0b5a4b
db5664e
 
47031d7
a4e24d4
02fd6bb
47031d7
a4e24d4
47031d7
 
da1009f
02fd6bb
 
 
a4e24d4
799409f
 
db5664e
da1009f
799409f
da1009f
 
 
2a73022
 
 
 
 
 
 
 
9a5d1e4
da1009f
9a5d1e4
02fd6bb
47031d7
 
02fd6bb
 
 
 
 
 
 
73ca5b8
 
 
 
 
 
 
 
 
1f1d641
ab616bd
 
 
73ca5b8
ab616bd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
94c7e5b
73ca5b8
 
 
 
 
ab616bd
73ca5b8
 
63fdbaa
02fd6bb
63fdbaa
 
 
 
 
 
 
 
 
 
 
 
 
02fd6bb
8f2f662
 
 
 
 
 
a4e24d4
73ca5b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eb5a3fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
 
 
 
 
 
 
 
ab616bd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
 
 
d0b5a4b
 
 
 
 
73ca5b8
 
 
 
 
 
 
d0b5a4b
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
 
 
d0b5a4b
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
 
 
 
d0b5a4b
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
 
 
d0b5a4b
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
 
 
 
d0b5a4b
 
 
 
 
 
 
 
 
 
73ca5b8
 
 
 
 
 
d0b5a4b
 
 
 
 
a4e24d4
02fd6bb
a4e24d4
 
 
 
63fdbaa
02fd6bb
63fdbaa
 
 
 
 
 
 
47031d7
a4e24d4
da1009f
 
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
import json
from pathlib import Path

import httpx
from typing import Optional, AsyncIterator, Dict, Any, Iterator, List, Callable
import logging
import asyncio
from litserve import LitAPI
from pydantic import BaseModel
from .utils import extract_json


class GenerationResponse(BaseModel):
    generated_text: str

class InferenceApi(LitAPI):
    def __init__(self, config: Dict[str, Any]):
        """Initialize the Inference API with configuration."""
        super().__init__()
        self.logger = logging.getLogger(__name__)
        self.logger.info("Initializing Inference API")
        self._device = None
        self.stream = False
        self.config = config
        self.llm_config = config.get('llm_server', {})

    def setup(self, device: Optional[str] = None):
        """Synchronous setup method required by LitAPI"""
        self._device = device
        self.logger.info(f"Inference API setup completed on device: {device}")
        return self  # It's common for setup methods to return self for chaining

    async def _get_client(self):
        """Get or create HTTP client as needed"""
        host = self.llm_config.get('host', 'localhost')
        port = self.llm_config.get('port', 8002)

        # Construct base URL, omitting port for HF spaces
        if 'hf.space' in host:
            base_url = f"https://{host}"
        else:
            base_url = f"http://{host}:{port}"

        return httpx.AsyncClient(
            base_url=base_url,
            timeout=float(self.llm_config.get('timeout', 60.0))
        )

    def _get_endpoint(self, endpoint_name: str) -> str:
        """Get full endpoint path including prefix"""
        endpoints = self.llm_config.get('endpoints', {})
        api_prefix = self.llm_config.get('api_prefix', '')
        endpoint = endpoints.get(endpoint_name, '')
        return f"{api_prefix}{endpoint}"

    async def _make_request(
            self,
            method: str,
            endpoint: str,
            *,
            params: Optional[Dict[str, Any]] = None,
            json: Optional[Dict[str, Any]] = None,
            stream: bool = False
    ) -> Any:
        """Make an authenticated request to the LLM Server."""
        base_url = self.llm_config.get('host', 'http://localhost:8001')
        full_endpoint = f"{base_url.rstrip('/')}/{self._get_endpoint(endpoint).lstrip('/')}"

        try:
            self.logger.info(f"Making {method} request to: {full_endpoint}")
            # Create client outside the with block for streaming
            client = await self._get_client()

            if stream:
                # For streaming, return both client and response context managers
                return client, client.stream(
                    method,
                    self._get_endpoint(endpoint),
                    params=params,
                    json=json
                )
            else:
                # For non-streaming, use context manager
                async with client as c:
                    response = await c.request(
                        method,
                        self._get_endpoint(endpoint),
                        params=params,
                        json=json
                    )
                    response.raise_for_status()
                    return response

        except Exception as e:
            self.logger.error(f"Error in request to {full_endpoint}: {str(e)}")
            raise

    def predict(self, x: str, **kwargs) -> Iterator[str]:
        """Non-async prediction method that yields results."""
        loop = asyncio.get_event_loop()
        async def async_gen():
            async for item in self._async_predict(x, **kwargs):
                yield item

        gen = async_gen()
        while True:
            try:
                yield loop.run_until_complete(gen.__anext__())
            except StopAsyncIteration:
                break

    async def _async_predict(self, x: str, **kwargs) -> AsyncIterator[str]:
        """Internal async prediction method."""
        if self.stream:
            async for chunk in self.generate_stream(x, **kwargs):
                yield chunk
        else:
            response = await self.generate_response(x, **kwargs)
            yield response

    async def generate_response(
            self,
            prompt: str,
            system_message: Optional[str] = None,
            max_new_tokens: Optional[int] = None
    ) -> str:
        """Generate a complete response by forwarding the request to the LLM Server."""
        self.logger.debug(f"Forwarding generation request for prompt: {prompt[:50]}...")

        try:
            response = await self._make_request(
                "POST",
                "generate",
                json={
                    "prompt": prompt,
                    "system_message": system_message,
                    "max_new_tokens": max_new_tokens
                }
            )
            data = response.json()
            return data["generated_text"]

        except Exception as e:
            self.logger.error(f"Error in generate_response: {str(e)}")
            raise

    async def structured_llm_query(
            self,
            template_name: str,
            input_text: str,
            additional_context: Optional[Dict[str, Any]] = None,
            pre_hooks: Optional[List[Callable]] = None,
            post_hooks: Optional[List[Callable]] = None
    ) -> Dict[str, Any]:
        """Execute a structured LLM query using a template."""
        template_path = Path(__file__).parent / "prompt_templates" / f"{template_name}.json"

        try:
            # Load and parse template
            with open(template_path) as f:
                template = json.load(f)

            # Apply pre-processing hooks
            processed_input = input_text
            if pre_hooks:
                for hook in pre_hooks:
                    processed_input = hook(processed_input)

            # Format the prompt with the context
            context = {"input_text": processed_input}
            if additional_context:
                context.update(additional_context)

            prompt = template["prompt_template"].format(**context)

            # Make the request to the LLM
            response = await self._make_request(
                "POST",
                "generate",
                json={
                    "prompt": prompt,
                    "system_message": template.get("system_message"),
                    "max_new_tokens": 1000
                }
            )

            # Extract JSON from response
            data = response.json()
            result = extract_json(data["generated_text"])

            # Apply any additional post-processing hooks
            if post_hooks:
                for hook in post_hooks:
                    result = hook(result)

            return result

        except FileNotFoundError:
            raise ValueError(f"Template {template_name} not found")
        except Exception as e:
            self.logger.error(f"Error in structured_llm_query: {str(e)}")
            raise

    async def expand_query(
            self,
            query: str,
            system_message: Optional[str] = None
    ) -> Dict[str, Any]:
        """Expand a query for RAG processing."""
        return await self.structured_llm_query(
            template_name="query_expansion",
            input_text=query,
            additional_context={"system_message": system_message} if system_message else None
        )

    async def rerank_chunks(
            self,
            query: str,
            chunks: List[str],
            system_message: Optional[str] = None
    ) -> Dict[str, Any]:
        """Rerank text chunks based on their relevance to the query."""
        # Format chunks as numbered list for better LLM processing
        formatted_chunks = "\n".join(f"{i+1}. {chunk}" for i, chunk in enumerate(chunks))

        return await self.structured_llm_query(
            template_name="chunk_rerank",
            input_text=query,
            additional_context={
                "chunks": formatted_chunks,
                "system_message": system_message
            }
        )


    async def generate_stream(
            self,
            prompt: str,
            system_message: Optional[str] = None,
            max_new_tokens: Optional[int] = None
    ) -> AsyncIterator[str]:
        """Generate a streaming response by forwarding the request to the LLM Server."""
        self.logger.debug(f"Forwarding streaming request for prompt: {prompt[:50]}...")

        try:
            client, stream_cm = await self._make_request(
                "POST",
                "generate_stream",
                json={
                    "prompt": prompt,
                    "system_message": system_message,
                    "max_new_tokens": max_new_tokens
                },
                stream=True
            )

            async with client:
                async with stream_cm as response:
                    async for chunk in response.aiter_text():
                        yield chunk

        except Exception as e:
            self.logger.error(f"Error in generate_stream: {str(e)}")
            raise

    async def generate_embedding(self, text: str) -> List[float]:
        """Generate embedding vector from input text."""
        self.logger.debug(f"Forwarding embedding request for text: {text[:50]}...")

        try:
            response = await self._make_request(
                "POST",
                "embedding",
                json={"text": text}
            )
            data = response.json()
            return data["embedding"]

        except Exception as e:
            self.logger.error(f"Error in generate_embedding: {str(e)}")
            raise

    async def check_system_status(self) -> Dict[str, Any]:
        """Check system status of the LLM Server."""
        self.logger.debug("Checking system status...")

        try:
            response = await self._make_request(
                "GET",
                "system_status"
            )
            return response.json()

        except Exception as e:
            self.logger.error(f"Error in check_system_status: {str(e)}")
            raise

    async def download_model(self, model_name: Optional[str] = None) -> Dict[str, str]:
        """Download model files from the LLM Server."""
        self.logger.debug(f"Forwarding model download request for: {model_name or 'default model'}")

        try:
            response = await self._make_request(
                "POST",
                "model_download",
                params={"model_name": model_name} if model_name else None
            )
            return response.json()

        except Exception as e:
            self.logger.error(f"Error in download_model: {str(e)}")
            raise

    async def validate_system(self) -> Dict[str, Any]:
        """Validate system configuration and setup."""
        self.logger.debug("Validating system configuration...")

        try:
            response = await self._make_request(
                "GET",
                "system_validate"
            )
            return response.json()

        except Exception as e:
            self.logger.error(f"Error in validate_system: {str(e)}")
            raise

    async def initialize_model(self, model_name: Optional[str] = None) -> Dict[str, Any]:
        """Initialize specified model or default model."""
        self.logger.debug(f"Initializing model: {model_name or 'default'}")

        try:
            response = await self._make_request(
                "POST",
                "model_initialize",
                params={"model_name": model_name} if model_name else None
            )
            return response.json()

        except Exception as e:
            self.logger.error(f"Error in initialize_model: {str(e)}")
            raise

    async def initialize_embedding_model(self, model_name: Optional[str] = None) -> Dict[str, Any]:
        """Initialize embedding model."""
        self.logger.debug(f"Initializing embedding model: {model_name or 'default'}")

        try:
            response = await self._make_request(
                "POST",
                "model_initialize_embedding",
                json={"model_name": model_name} if model_name else {}
            )
            return response.json()

        except Exception as e:
            self.logger.error(f"Error in initialize_embedding_model: {str(e)}")
            raise

    def decode_request(self, request: Any, **kwargs) -> str:
        """Convert the request payload to input format."""
        if isinstance(request, dict) and "prompt" in request:
            return request["prompt"]
        return request

    def encode_response(self, output: Iterator[str], **kwargs) -> Dict[str, Any]:
        """Convert the model output to a response payload."""
        if self.stream:
            return {"generated_text": output}
        try:
            result = next(output)
            return {"generated_text": result}
        except StopIteration:
            return {"generated_text": ""}

    async def cleanup(self):
        """Cleanup method - no longer needed as clients are created per-request"""
        pass