File size: 20,665 Bytes
94ddc94
 
 
 
 
 
 
 
093632f
 
83d1ea1
 
 
52ee99e
 
94ddc94
 
 
 
093632f
94ddc94
093632f
 
 
 
 
83d1ea1
 
093632f
 
 
 
6b7de2a
 
04f2d25
 
83d1ea1
 
52ee99e
 
093632f
 
 
 
94ddc94
 
 
093632f
94ddc94
 
093632f
32585a6
093632f
 
52ee99e
04f2d25
 
52ee99e
 
04f2d25
6b7de2a
 
 
093632f
83d1ea1
 
 
 
 
32585a6
83d1ea1
32585a6
04f2d25
7ba4670
04f2d25
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
18fc113
 
 
 
 
04f2d25
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9b1726d
04f2d25
52ee99e
7ba4670
52ee99e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9b1726d
18fc113
 
 
 
 
 
 
 
 
 
 
9b1726d
52ee99e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a4d8b4b
9b1726d
a4d8b4b
04f2d25
 
7ba4670
32585a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
18fc113
32585a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
18fc113
 
 
 
 
32585a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9b1726d
32585a6
093632f
 
 
 
 
 
 
 
 
 
 
9023f08
093632f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86ff876
 
093632f
 
 
 
 
 
 
9b1726d
093632f
7ba4670
83d1ea1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
18fc113
 
 
83d1ea1
 
18fc113
 
 
 
83d1ea1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9b1726d
83d1ea1
093632f
 
 
 
18fc113
 
093632f
 
18fc113
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
093632f
18fc113
 
 
 
 
 
 
 
 
 
 
 
 
093632f
83d1ea1
d59e3c0
 
 
 
 
 
 
 
 
 
 
 
83d1ea1
 
d59e3c0
83d1ea1
d59e3c0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
83d1ea1
093632f
 
 
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
"""
This class contains multiple LLMs and handles LLMs response
"""

import json
import time
from openai import OpenAI
import openai
import torch
import re
import anthropic
import os
import streamlit as st
from google.genai import types
from google import genai



class LLM:
    def __init__(self, Core):
        self.Core = Core
        self.model = None
        self.model_type = "openai" # valid values -> ["openai", "ollama"]
        self.client = None
        self.connect_to_llm()

    def get_credential(self, key):
        return os.getenv(key) or st.secrets.get(key)

    def get_response(self, prompt, instructions):
        if self.model_type == "openai":
            response = self.get_message_openai(prompt, instructions)
        # elif self.model_type == "ollama":
        #     response = self.get_message_ollama(prompt, instructions)
        elif self.model_type == "inference":
            response = self.get_message_inference(prompt, instructions)
        elif self.model_type == "claude":
            response = self.get_message_claude(prompt, instructions)
        elif self.model_type == "google":
            response = self.get_message_google(prompt, instructions)
        else:
            raise f"Invalid model type : {self.model_type}"

        return response

    def connect_to_llm(self):
        """
        connect to selected llm -> ollama or openai connection
        :return:
        """

        if self.Core.model in self.Core.config_file["openai_models"]:
            self.model_type = "openai"

        elif self.Core.model in self.Core.config_file["inference_models"]:
            self.model_type = "inference"

        elif self.Core.model in self.Core.config_file["google_models"]:
            self.model_type = "google"

        # elif self.Core.model in self.Core.config_file["ollama_models"]:
        #     self.model_type = "ollama"
        #     self.client = ollama.Client()

        elif self.Core.model in self.Core.config_file["claude_models"]:
            self.model_type = "claude"
            self.client = anthropic.Anthropic(
                api_key=self.get_credential('claude_api_key'),
            )

        self.model = self.Core.model

    # ==============================================================
    def get_message_inference(self, prompt, instructions, max_retries=6):
        """
        sending the prompt to openai LLM and get back the response
        """

        api_key = self.get_credential('inference_api_key')
        client = OpenAI(
            base_url="https://api.inference.net/v1",
            api_key=api_key,
        )

        for attempt in range(max_retries):
            try:
                if self.Core.reasoning_model:
                    response = client.chat.completions.create(
                        model=self.Core.model,
                        response_format={"type": "json_object"},
                        messages=[
                            {"role": "system", "content": instructions},
                            {"role": "user", "content": prompt}
                        ],
                        reasoning_effort="medium",
                        n=1,
                    )

                else:
                    response = client.chat.completions.create(
                        model=self.Core.model,
                        response_format={"type": "json_object"},
                        messages=[
                            {"role": "system", "content": instructions},
                            {"role": "user", "content": prompt}
                        ],
                        n=1,
                        temperature=self.Core.temperature
                    )

                tokens = {
                    'prompt_tokens': response.usage.prompt_tokens,
                    'completion_tokens': response.usage.completion_tokens,
                    'total_tokens': response.usage.total_tokens
                }

                # validating the JSON
                self.Core.total_tokens['prompt_tokens'] += tokens['prompt_tokens']
                self.Core.total_tokens['completion_tokens'] += tokens['completion_tokens']
                self.Core.temp_token_counter += tokens['total_tokens']

                try:
                    content = response.choices[0].message.content

                    # Extract JSON code block

                    output = json.loads(content)

                    if 'message' not in output or 'header' not in output:
                        print(f"'message' or 'header' is missing in response on attempt {attempt + 1}. Retrying...")
                        continue  # Continue to next attempt

                    else:
                        if len(output["header"].strip()) > self.Core.config_file["header_limit"] or len(
                                output["message"].strip()) > self.Core.config_file["message_limit"]:
                            print(
                                f"'header' or 'message' is more than specified characters in response on attempt {attempt + 1}. Retrying...")
                            continue

                    return output

                except json.JSONDecodeError:
                    print(f"Invalid JSON from LLM on attempt {attempt + 1}. Retrying...")

            except openai.APIConnectionError as e:
                print("The server could not be reached")
                print(e.__cause__)  # an underlying Exception, likely raised within httpx.
            except openai.RateLimitError as e:
                print("A 429 status code was received; we should back off a bit.")
            except openai.APIStatusError as e:
                print("Another non-200-range status code was received")
                print(e.status_code)
                print(e.response)

        print("Max retries exceeded. Returning empty response.")
        return None

    # =========================================================================
    def get_message_google(self, prompt, instructions, max_retries=6):

        client = genai.Client(api_key=self.get_credential("Google_API"))

        for attempt in range(max_retries):
            try:
                response = client.models.generate_content(
                    model=self.Core.model,
                    contents=prompt,
                    config=types.GenerateContentConfig(
                        thinking_config=types.ThinkingConfig(thinking_budget=0),
                        system_instruction=instructions,
                        temperature=self.Core.temperature,
                        response_mime_type="application/json"
                    ))

                # output = json.loads(str(response.text))
                tokens = {
                    'prompt_tokens': response.usage_metadata.prompt_token_count,
                    'completion_tokens': response.usage_metadata.candidates_token_count,
                    'total_tokens': response.usage_metadata.total_token_count
                }

                # validating the JSON
                self.Core.total_tokens['prompt_tokens'] += tokens['prompt_tokens']
                self.Core.total_tokens['completion_tokens'] += tokens['completion_tokens']
                self.Core.temp_token_counter += tokens['total_tokens']

                output = self.preprocess_and_parse_json(response.text)

                if 'message' not in output or 'header' not in output:
                    print(f"'message' or 'header' is missing in response on attempt {attempt + 1}. Retrying...")
                    continue  # Continue to next attempt

                else:
                    if len(output["header"].strip()) > self.Core.config_file["header_limit"] or len(
                            output["message"].strip()) > self.Core.config_file["message_limit"]:
                        print(
                            f"'header' or 'message' is more than specified characters in response on attempt {attempt + 1}. Retrying...")
                        continue
                return output

            except json.JSONDecodeError:
                print(f"Invalid JSON from LLM on attempt {attempt + 1}. Retrying...")
            except Exception as e:
                print(f"Error in attempt {attempt}: {e}")

        print("Max retries exceeded. Returning empty response.")
        return None

    # =========================================================================

    def get_message_openai(self, prompt, instructions, max_retries=5):
        """
        sending the prompt to openai LLM and get back the response
        """

        openai.api_key = self.Core.api_key
        client = OpenAI(api_key=self.Core.api_key)

        for attempt in range(max_retries):
            try:
                if self.Core.reasoning_model:
                    response = client.chat.completions.create(
                        model=self.Core.model,
                        response_format={"type": "json_object"},
                        messages=[
                            {"role": "system", "content": instructions},
                            {"role": "user", "content": prompt}
                        ],
                        reasoning_effort="minimal",
                        n=1,
                    )

                else:
                    response = client.chat.completions.create(
                        model=self.Core.model,
                        response_format={"type": "json_object"},
                        messages=[
                            {"role": "system", "content": instructions},
                            {"role": "user", "content": prompt}
                        ],
                        n=1,
                        temperature=self.Core.temperature
                    )

                tokens = {
                    'prompt_tokens': response.usage.prompt_tokens,
                    'completion_tokens': response.usage.completion_tokens,
                    'total_tokens': response.usage.total_tokens
                }

                # validating the JSON
                self.Core.total_tokens['prompt_tokens'] += tokens['prompt_tokens']
                self.Core.total_tokens['completion_tokens'] += tokens['completion_tokens']
                self.Core.temp_token_counter += tokens['total_tokens']

                try:
                    content = response.choices[0].message.content

                    # Extract JSON code block

                    output = json.loads(content)

                    if 'message' not in output or 'header' not in output:
                        print(f"'message' or 'header' is missing in response on attempt {attempt + 1}. Retrying...")
                        continue  # Continue to next attempt

                    else:
                        if len(output["header"].strip()) > self.Core.config_file["header_limit"] or len(
                                output["message"].strip()) > self.Core.config_file["message_limit"]:
                            print(
                                f"'header' or 'message' is more than specified characters in response on attempt {attempt + 1}. Retrying...")
                            continue

                    return output

                except json.JSONDecodeError:
                    print(f"Invalid JSON from LLM on attempt {attempt + 1}. Retrying...")

            except openai.APIConnectionError as e:
                print("The server could not be reached")
                print(e.__cause__)  # an underlying Exception, likely raised within httpx.
            except openai.RateLimitError as e:
                print("A 429 status code was received; we should back off a bit.")
            except openai.APIStatusError as e:
                print("Another non-200-range status code was received")
                print(e.status_code)
                print(e.response)

        print("Max retries exceeded. Returning empty response.")
        return None

    # ======================================================================

    def get_message_ollama(self, prompt, instructions, max_retries=10):
        """
        Send the prompt to the LLM and get back the response.
        Includes handling for GPU memory issues by clearing cache and waiting before retry.
        """
        prompt = instructions + "\n \n" + prompt
        for attempt in range(max_retries):
            try:
                # Try generating the response
                response = self.client.generate(model=self.model, prompt=prompt)
            except Exception as e:
                # This catches errors like the connection being forcibly closed
                print(f"Error on attempt {attempt + 1}: {e}.")
                try:
                    # Clear GPU cache if you're using PyTorch; this may help free up memory
                    torch.cuda.empty_cache()
                    print("Cleared GPU cache.")
                except Exception as cache_err:
                    print("Failed to clear GPU cache:", cache_err)
                # Wait a bit before retrying to allow memory to recover
                time.sleep(2)
                continue

            try:
                tokens = {
                    'prompt_tokens': 0,
                    'completion_tokens': 0,
                    'total_tokens': 0
                }

                try:
                    output = self.preprocess_and_parse_json(response.response)
                    if output is None:
                        continue

                    if 'message' not in output or 'header' not in output:
                        print(f"'message' or 'header' is missing in response on attempt {attempt + 1}. Retrying...")
                        continue  # Continue to next attempt

                    else:
                        if len(output["header"].strip()) > self.Core.config_file["header_limit"] or len(
                                output["message"].strip()) > self.Core.config_file["message_limit"]:
                            print(
                                f"'header' or 'message' is more than specified characters in response on attempt {attempt + 1}. Retrying...")
                            continue
                        else:
                            return output

                except json.JSONDecodeError:
                    print(f"Invalid JSON from LLM on attempt {attempt + 1}. Retrying...")
            except Exception as parse_error:
                print("Error processing output:", parse_error)

        print("Max retries exceeded. Returning empty response.")
        return None

    def get_message_claude(self, prompt, instructions, max_retries=6):
        """
        send prompt to claude LLM and get back the response
        :param prompt:
        :param instructions:
        :return:
        """


        for attempt in range(max_retries):
            try:

                message = self.client.messages.create(
                    model=self.model,
                    max_tokens=4096,
                    system = instructions,
                    messages=[
                        {"role": "user", "content": prompt + "\nHere is the JSON requested:\n"}
                    ],
                    temperature=self.Core.temperature
                )
                # Try generating the response
                response = message.content[0].text

                tokens = {
                    'prompt_tokens': message.usage.input_tokens,
                    'completion_tokens': message.usage.output_tokens,
                    'total_tokens': message.usage.output_tokens + message.usage.input_tokens
                }

                self.Core.total_tokens['prompt_tokens'] += tokens['prompt_tokens']
                self.Core.total_tokens['completion_tokens'] += tokens['completion_tokens']
                self.Core.temp_token_counter += tokens['total_tokens']

                try:
                    output = self.preprocess_and_parse_json_claude(response)
                    if output is None:
                        continue

                    if 'message' not in output or 'header' not in output:
                        print(f"'message' or 'header' is missing in response on attempt {attempt + 1}. Retrying...")
                        continue  # Continue to next attempt

                    else:
                        if len(output["header"].strip()) > self.Core.config_file["header_limit"] or len(
                                output["message"].strip()) > self.Core.config_file["message_limit"]:
                            print(
                                f"'header' or 'message' is more than specified characters in response on attempt {attempt + 1}. Retrying...")
                            continue
                        else:
                            return output

                except json.JSONDecodeError:
                    print(f"Invalid JSON from LLM on attempt {attempt + 1}. Retrying...")
            except Exception as parse_error:
                print("Error processing output:", parse_error)

        print("Max retries exceeded. Returning empty response.")
        return None

    # ======================================================================

    def preprocess_and_parse_json(self, response: str):
        """
        Remove <think> blocks, extract JSON (from ```json fences or first {...} block),
        and parse. Includes a repair pass to handle common LLM issues like trailing commas.
        """

        def extract_json(text: str) -> str:
            # Remove <think>...</think>
            text = re.sub(r'<think>.*?</think>', '', text, flags=re.DOTALL).strip()

            # Prefer fenced code if present
            fence = re.search(r'```(?:json)?(.*?)```', text, flags=re.DOTALL | re.IGNORECASE)
            if fence:
                return fence.group(1).strip()

            # Otherwise, grab the first {...} block
            brace = re.search(r'\{.*\}', text, flags=re.DOTALL)
            return brace.group(0).strip() if brace else text.strip()

        def normalize_quotes(text: str) -> str:
            return (text
                    .replace('\ufeff', '')  # strip BOM if present
                    .replace('“', '"').replace('”', '"')
                    .replace('‘', "'").replace('’', "'"))

        def strip_comments(text: str) -> str:
            # Remove // line comments and /* block comments */
            text = re.sub(r'//.*?$', '', text, flags=re.MULTILINE)
            text = re.sub(r'/\*.*?\*/', '', text, flags=re.DOTALL)
            return text

        def remove_trailing_commas(text: str) -> str:
            # Remove commas before } or ]
            return re.sub(r',(\s*[}\]])', r'\1', text)

        raw = extract_json(response)
        raw = normalize_quotes(raw)

        try:
            return json.loads(raw)
        except json.JSONDecodeError:
            # Repair pass
            repaired = strip_comments(raw)
            repaired = remove_trailing_commas(repaired)
            repaired = repaired.strip()

            try:
                return json.loads(repaired)
            except json.JSONDecodeError as e:
                print(f"Failed to parse JSON: {e}")
                # print('Offending text:', repaired)
                return None

    # ===============================================================
    # def preprocess_and_parse_json_claude(self, response: str):
    #     """
    #     process claude response and extract JSON
    #     :param response:
    #     :return:
    #     """
    #     json_start = response.index("{")
    #     json_end = response.rfind("}")
    #     parsed_response = json.loads(response[json_start:json_end + 1])
    #     return parsed_response
    #

    def preprocess_and_parse_json_claude(self, response: str):
        """
        Process Claude response and extract JSON content safely
        """
        try:
            json_start = response.index("{")
            json_end = response.rfind("}")
            json_string = response[json_start:json_end + 1]

            parsed_response = json.loads(json_string)

            if not isinstance(parsed_response, dict):
                raise ValueError(f"Parsed response is not a dict: {parsed_response}")

            return parsed_response

        except ValueError as ve:
            raise ValueError(f"Could not extract JSON from Claude response: {ve}\nOriginal response: {response}")
        except json.JSONDecodeError as je:
            raise ValueError(f"Failed to parse JSON from string: {json_string}\nError: {je}")