|
1
|
"""AWS Bedrock provider implementation.""" |
|
2
|
|
|
3
|
import base64 |
|
4
|
import json |
|
5
|
import logging |
|
6
|
import os |
|
7
|
from pathlib import Path |
|
8
|
from typing import Optional |
|
9
|
|
|
10
|
from dotenv import load_dotenv |
|
11
|
|
|
12
|
from video_processor.providers.base import BaseProvider, ModelInfo, ProviderRegistry |
|
13
|
|
|
14
|
load_dotenv() |
|
15
|
logger = logging.getLogger(__name__) |
|
16
|
|
|
17
|
# Curated list of popular Bedrock models |
|
18
|
_BEDROCK_MODELS = [ |
|
19
|
ModelInfo( |
|
20
|
id="anthropic.claude-3-5-sonnet-20241022-v2:0", |
|
21
|
provider="bedrock", |
|
22
|
display_name="Claude 3.5 Sonnet v2", |
|
23
|
capabilities=["chat", "vision"], |
|
24
|
), |
|
25
|
ModelInfo( |
|
26
|
id="anthropic.claude-3-sonnet-20240229-v1:0", |
|
27
|
provider="bedrock", |
|
28
|
display_name="Claude 3 Sonnet", |
|
29
|
capabilities=["chat", "vision"], |
|
30
|
), |
|
31
|
ModelInfo( |
|
32
|
id="anthropic.claude-3-haiku-20240307-v1:0", |
|
33
|
provider="bedrock", |
|
34
|
display_name="Claude 3 Haiku", |
|
35
|
capabilities=["chat", "vision"], |
|
36
|
), |
|
37
|
ModelInfo( |
|
38
|
id="amazon.titan-text-express-v1", |
|
39
|
provider="bedrock", |
|
40
|
display_name="Amazon Titan Text Express", |
|
41
|
capabilities=["chat"], |
|
42
|
), |
|
43
|
ModelInfo( |
|
44
|
id="meta.llama3-70b-instruct-v1:0", |
|
45
|
provider="bedrock", |
|
46
|
display_name="Llama 3 70B Instruct", |
|
47
|
capabilities=["chat"], |
|
48
|
), |
|
49
|
ModelInfo( |
|
50
|
id="mistral.mistral-large-2402-v1:0", |
|
51
|
provider="bedrock", |
|
52
|
display_name="Mistral Large", |
|
53
|
capabilities=["chat"], |
|
54
|
), |
|
55
|
] |
|
56
|
|
|
57
|
|
|
58
|
class BedrockProvider(BaseProvider): |
|
59
|
"""AWS Bedrock provider using boto3.""" |
|
60
|
|
|
61
|
provider_name = "bedrock" |
|
62
|
|
|
63
|
def __init__( |
|
64
|
self, |
|
65
|
aws_access_key_id: Optional[str] = None, |
|
66
|
aws_secret_access_key: Optional[str] = None, |
|
67
|
region_name: Optional[str] = None, |
|
68
|
): |
|
69
|
try: |
|
70
|
import boto3 |
|
71
|
except ImportError: |
|
72
|
raise ImportError("boto3 package not installed. Install with: pip install boto3") |
|
73
|
|
|
74
|
self._boto3 = boto3 |
|
75
|
self._region = region_name or os.getenv("AWS_DEFAULT_REGION", "us-east-1") |
|
76
|
self._client = boto3.client( |
|
77
|
"bedrock-runtime", |
|
78
|
aws_access_key_id=aws_access_key_id or os.getenv("AWS_ACCESS_KEY_ID"), |
|
79
|
aws_secret_access_key=aws_secret_access_key or os.getenv("AWS_SECRET_ACCESS_KEY"), |
|
80
|
region_name=self._region, |
|
81
|
) |
|
82
|
self._last_usage = {} |
|
83
|
|
|
84
|
def chat( |
|
85
|
self, |
|
86
|
messages: list[dict], |
|
87
|
max_tokens: int = 4096, |
|
88
|
temperature: float = 0.7, |
|
89
|
model: Optional[str] = None, |
|
90
|
) -> str: |
|
91
|
model = model or "anthropic.claude-3-sonnet-20240229-v1:0" |
|
92
|
# Strip bedrock/ prefix if present |
|
93
|
if model.startswith("bedrock/"): |
|
94
|
model = model[len("bedrock/") :] |
|
95
|
|
|
96
|
body = json.dumps( |
|
97
|
{ |
|
98
|
"anthropic_version": "bedrock-2023-05-31", |
|
99
|
"max_tokens": max_tokens, |
|
100
|
"temperature": temperature, |
|
101
|
"messages": messages, |
|
102
|
} |
|
103
|
) |
|
104
|
|
|
105
|
response = self._client.invoke_model( |
|
106
|
modelId=model, |
|
107
|
contentType="application/json", |
|
108
|
accept="application/json", |
|
109
|
body=body, |
|
110
|
) |
|
111
|
|
|
112
|
result = json.loads(response["body"].read()) |
|
113
|
self._last_usage = { |
|
114
|
"input_tokens": result.get("usage", {}).get("input_tokens", 0), |
|
115
|
"output_tokens": result.get("usage", {}).get("output_tokens", 0), |
|
116
|
} |
|
117
|
return result.get("content", [{}])[0].get("text", "") |
|
118
|
|
|
119
|
def analyze_image( |
|
120
|
self, |
|
121
|
image_bytes: bytes, |
|
122
|
prompt: str, |
|
123
|
max_tokens: int = 4096, |
|
124
|
model: Optional[str] = None, |
|
125
|
) -> str: |
|
126
|
model = model or "anthropic.claude-3-sonnet-20240229-v1:0" |
|
127
|
if model.startswith("bedrock/"): |
|
128
|
model = model[len("bedrock/") :] |
|
129
|
|
|
130
|
b64 = base64.b64encode(image_bytes).decode() |
|
131
|
body = json.dumps( |
|
132
|
{ |
|
133
|
"anthropic_version": "bedrock-2023-05-31", |
|
134
|
"max_tokens": max_tokens, |
|
135
|
"messages": [ |
|
136
|
{ |
|
137
|
"role": "user", |
|
138
|
"content": [ |
|
139
|
{ |
|
140
|
"type": "image", |
|
141
|
"source": { |
|
142
|
"type": "base64", |
|
143
|
"media_type": "image/jpeg", |
|
144
|
"data": b64, |
|
145
|
}, |
|
146
|
}, |
|
147
|
{"type": "text", "text": prompt}, |
|
148
|
], |
|
149
|
} |
|
150
|
], |
|
151
|
} |
|
152
|
) |
|
153
|
|
|
154
|
response = self._client.invoke_model( |
|
155
|
modelId=model, |
|
156
|
contentType="application/json", |
|
157
|
accept="application/json", |
|
158
|
body=body, |
|
159
|
) |
|
160
|
|
|
161
|
result = json.loads(response["body"].read()) |
|
162
|
self._last_usage = { |
|
163
|
"input_tokens": result.get("usage", {}).get("input_tokens", 0), |
|
164
|
"output_tokens": result.get("usage", {}).get("output_tokens", 0), |
|
165
|
} |
|
166
|
return result.get("content", [{}])[0].get("text", "") |
|
167
|
|
|
168
|
def transcribe_audio( |
|
169
|
self, |
|
170
|
audio_path: str | Path, |
|
171
|
language: Optional[str] = None, |
|
172
|
model: Optional[str] = None, |
|
173
|
) -> dict: |
|
174
|
raise NotImplementedError( |
|
175
|
"AWS Bedrock does not support audio transcription directly. " |
|
176
|
"Use Amazon Transcribe or another provider for transcription." |
|
177
|
) |
|
178
|
|
|
179
|
def list_models(self) -> list[ModelInfo]: |
|
180
|
return list(_BEDROCK_MODELS) |
|
181
|
|
|
182
|
|
|
183
|
ProviderRegistry.register( |
|
184
|
name="bedrock", |
|
185
|
provider_class=BedrockProvider, |
|
186
|
env_var="AWS_ACCESS_KEY_ID", |
|
187
|
model_prefixes=["bedrock/"], |
|
188
|
default_models={ |
|
189
|
"chat": "anthropic.claude-3-sonnet-20240229-v1:0", |
|
190
|
"vision": "anthropic.claude-3-sonnet-20240229-v1:0", |
|
191
|
"audio": "", |
|
192
|
}, |
|
193
|
) |
|
194
|
|