Commit Graph

74 Commits

Author SHA1 Message Date
nai-degen 26dc79c8f1 fixes broken AWS rate limit backoff 2023-10-24 09:19:46 -05:00
nai-degen 89e9b67f3f fixes AWS mid-stream rate limits not actually marking key as rate-limited 2023-10-23 22:47:29 -05:00
nai-degen 52ec2ec265 fixes blank AWS responses due to reqs sometimes using wrong handler 2023-10-23 22:23:06 -05:00
nai-degen 8bd2f749c1 reduces logging severity of prompt validation errors 2023-10-23 20:30:27 -05:00
nai-degen 3f7e50f87e follow-up 'fixes empty AWS streaming responses when under heavy load' 2023-10-15 00:06:38 -05:00
nai-degen f6cfc6e882 fixes empty AWS streaming responses when under heavy load 2023-10-15 00:05:36 -05:00
nai-degen af4d8dae40 changes default AMZ_HOST to bedrock-runtime.region.amazonaws.com 2023-10-12 15:39:06 -05:00
nai-degen 12f78fa1f2 exempts 'special' role from rate limiting 2023-10-06 20:29:28 -05:00
nai-degen 5033d00444 improves clarity of errors sent back to streaming clients 2023-10-03 19:45:15 -05:00
nai-degen ba0b20617e ensures AWS always uses anthropic-version 2023-06-01 parser 2023-10-03 19:43:30 -05:00
khanon ecf897e685 Refactor handleStreamingResponse to make it less shit (khanon/oai-reverse-proxy!46) 2023-10-03 06:14:19 +00:00
nai-degen ede274c117 disables AWS key on AccessDeniedException 2023-10-02 11:18:08 -05:00
nai-degen 0837c89a42 fixes incorrect context size limit for aws claude v1 2023-10-02 03:53:04 -05:00
nai-degen f67560a17b refactors proxy routing 2023-10-01 12:12:28 -05:00
nai-degen e13361a323 removes dead koboldai code 2023-10-01 11:27:11 -05:00
khanon fa4bf468d2 Implement AWS Bedrock support (khanon/oai-reverse-proxy!45) 2023-10-01 01:40:18 +00:00
nai-degen 7e681a7bef strips OAI request parameters when translating to Claude format 2023-09-29 03:01:39 -05:00
nai-degen 1b0106a1ea strips reverse proxy originating IP headers 2023-09-29 03:00:55 -05:00
nai-degen f5521aa6c3 prevents selecting trial keys for embeddings requests due to rate limits 2023-09-26 01:26:07 -05:00
nai-degen f8b480f4c2 adds support for proxying text-embedding-ada-002 requests 2023-09-26 00:58:38 -05:00
khanon 35b44e1c6b fixes issue with OpenAIV1ChatCompletionSchema and PaLM compat 2023-09-24 10:48:56 +00:00
nai-degen 075e415343 makes incoming model name validation less strict for PaLM endpoint 2023-09-20 23:55:53 -05:00
khanon 35a6c393ed Add support for Google PaLM and OpenAI Turbo Instruct (khanon/oai-reverse-proxy!44) 2023-09-19 23:13:08 +00:00
nai-degen 5e57dbb8f1 attempts to improve compatibility with BetterGPT frontend 2023-09-16 11:04:40 -05:00
nai-degen 7b3d6efb02 reverts anthropic-version change as it breaks some frontends 2023-09-07 22:01:19 -05:00
nai-degen 63542bfabb adds anthropic-version header in all cases 2023-09-07 20:23:34 -05:00
khanon f05e196994 Refactor project structure and add user self-serve UI (khanon/oai-reverse-proxy!41) 2023-09-02 19:36:44 +00:00
nai-degen 2c0a659b2d adds token consumption stats to infopage 2023-08-30 20:40:40 -05:00
khanon 4d781e1720 Add GPT-4-32k support (khanon/oai-reverse-proxy!39) 2023-08-29 22:56:54 +00:00
khanon 6833736392 Clone keys assigned to multiple organizations (khanon/oai-reverse-proxy!38) 2023-08-28 21:11:49 +00:00
khanon cb780e85da Per-user token quotas and automatic quota refreshing (khanon/oai-reverse-proxy!37) 2023-08-28 19:33:14 +00:00
nai-degen 6bb67281d9 removes QUEUE_MODE config (now always enabled) 2023-08-09 18:29:34 -05:00
nai-degen d1d83b41fa uses accurate Claude tokenization 2023-08-08 17:29:36 -05:00
nai-degen 125bbe6441 fixes issue with writeErrorResponse 2023-08-04 13:49:11 -05:00
nai-degen 00346360af fixes turbo-16k incompatibility 2023-07-23 20:13:38 -05:00
khanon 56a4902599 Add tokenizers and configurable context size limits (khanon/oai-reverse-proxy!28) 2023-07-22 00:11:32 +00:00
khanon aa5380d2ef Rework OpenAIKeyChecker to remove usage tracking and test all keys for liveness (khanon/oai-reverse-proxy!29) 2023-07-21 04:00:12 +00:00
breathingmanually 576423d1f8 Fix JSON parse exception when Claude finishes streaming (khanon/oai-reverse-proxy!25) 2023-07-20 01:57:50 +00:00
nai-degen d7a4829d13 handles keys which have been banned (but not revoked) by openai 2023-07-19 10:28:38 -05:00
nai-degen efa1b03570 uses claude-v1 by default as anthropic seems to be turning off v1.2 2023-07-19 01:48:57 -05:00
nai-degen 6598b4df0d requests old version of Anthropic API due to breaking SSE changes 2023-06-24 14:50:48 -05:00
nai-degen c8b3238398 reorganizes origin header middleware 2023-06-24 14:25:01 -05:00
nai-degen 602931bf7f removes origin/referer headers from proxied request 2023-06-23 00:08:09 -05:00
khanon 43359779e7 Implements more robust anti-zoomer functionality (khanon/oai-reverse-proxy!24) 2023-06-14 04:05:51 +00:00
khanon 96cf4a0e2d Automatically add Anthropic "\n\nHuman:" preamble when necessary (khanon/oai-reverse-proxy!23) 2023-06-09 07:16:40 +00:00
khanon 120040c028 Separate MAX_OUTPUT_TOKENS config for OpenAI/Claude (khanon/oai-reverse-proxy!21) 2023-06-08 00:58:57 +00:00
khanon 7b5e612921 Workaround for Anthropic requiring some prompts to start with Human: (khanon/oai-reverse-proxy!20) 2023-06-05 21:55:59 +00:00
khanon cd754bd601 Allow overriding default Claude model selection (khanon/oai-reverse-proxy!19) 2023-06-05 03:12:34 +00:00
nai-degen 38ff0e3f3b fixes prompt logging for streamed Claude requests 2023-06-04 12:16:29 -05:00
khanon dae1262f7a Refactor request middleware (khanon/oai-reverse-proxy!18) 2023-06-02 04:03:16 +00:00