Krrish Dholakia
124aae7579
fix(factory.py): add mistral config to known tokenizer configs
2024-04-18 17:11:52 -07:00
David Manouchehri
e22f22e0a9
(feat) - Add seed to Cohere Chat.
2024-04-18 20:57:06 +00:00
Krrish Dholakia
77cb75bf2d
fix(vertex_ai.py): fix faulty async completion function call tool check
2024-04-17 16:30:28 -07:00
Krrish Dholakia
72d7c36c76
refactor(utils.py): make it clearer how vertex ai params are handled '
...
'
2024-04-17 16:20:56 -07:00
Krrish Dholakia
6d508468ef
fix(vertex_ai.py): accept credentials as a json string
2024-04-16 17:34:25 -07:00
Krish Dholakia
b46e9058df
Merge pull request #3019 from ligaz/fix-anthropic-system-msg-handling
...
Fix Anthropic system message handling
2024-04-16 07:18:47 -07:00
Ishaan Jaff
2fbd800404
Merge pull request #3054 from BerriAI/litellm_supports_vision
...
[Feat] view models that `supports_vision`
2024-04-15 18:31:32 -07:00
Ishaan Jaff
1d96a54c34
fix - support base 64 image conversion for all gemini model
2024-04-15 18:18:55 -07:00
Krrish Dholakia
8c3c45fbb5
fix(vertex_ai_anthropic.py): set vertex_credentials for vertex ai anthropic calls
...
allows setting vertex credentials as a json string for vertex ai anthropic calls
2024-04-15 14:16:28 -07:00
Krrish Dholakia
3d645f95a5
fix(main.py): accept vertex service account credentials as json string
...
allows us to dynamically set vertex ai credentials
2024-04-15 13:28:59 -07:00
Krrish Dholakia
7df0626488
fix(anthropic_text.py): return process_response value
2024-04-15 08:51:26 -07:00
Krrish Dholakia
1cd0551a1e
fix(anthropic_text.py): add support for async text completion calls
2024-04-15 08:15:00 -07:00
Stefan Dobrev
b3b52dbe32
Fix Anthropic system message handling
2024-04-14 16:23:59 +03:00
Krrish Dholakia
ec72202d56
fix(gemini.py): log system prompt in verbose output
2024-04-11 23:15:58 -07:00
Krrish Dholakia
7a829f805d
fix(vertex_ai.py): check if 'response_mime_type' in generation config before passing it in
2024-04-11 23:10:59 -07:00
David Manouchehri
ad8d35a378
(feat) - Extreme dirty hack for response_mime_type in Vertex AI.
2024-04-12 00:03:31 +00:00
David Manouchehri
c6b866925f
(feat) - Dirty hack to get response_mime_type working before it's released in the Python SDK.
2024-04-12 00:03:30 +00:00
David Manouchehri
cc71ca3166
(feat) - Add support for JSON mode in Vertex AI
2024-04-12 00:03:29 +00:00
Ishaan Jaff
db0e63532b
Merge pull request #2892 from gregnwosu/patch-1
...
created defaults for response["eval_count"]
2024-04-11 13:40:32 -07:00
Ishaan Jaff
850241ada1
fix - stop using imghdr
2024-04-11 09:32:42 -07:00
Ishaan Jaff
9ee90a6bc3
fix cleanup
2024-04-10 21:28:04 -07:00
Ishaan Jaff
57fdfa0a5e
fix use ImmutableSandboxedEnvironment
2024-04-10 21:26:50 -07:00
Ishaan Jaff
f2816b3fcd
fix jinja2 use ImmutableSandboxedEnvironment
2024-04-10 20:30:13 -07:00
Ishaan Jaff
366a324735
fix - security report / vuln
2024-04-10 15:42:17 -07:00
Krrish Dholakia
76a298ec81
fix(replicate.py): handle empty response - replicate
2024-04-09 21:57:25 -07:00
Krrish Dholakia
57ea6a1c4f
feat(gemini.py): support google-genai system instruction
2024-04-09 17:23:14 -07:00
Gregory Nwosu
0609da96e3
created defaults for response["eval_count"]
...
there is no way in litellm to disable the cache in ollama that is removing the eval_count response keys from the json.
This PR allows the code to create sensible defaults for when the response is empty
see
- https://github.com/ollama/ollama/issues/1573
- https://github.com/ollama/ollama/issues/2023
2024-04-08 02:03:54 +01:00
Ishaan Jaff
494d71b063
add exit and aenter
2024-04-06 19:27:26 -07:00
Ishaan Jaff
833a64455f
ci/cd run async handler
2024-04-06 19:16:27 -07:00
Ishaan Jaff
1dc5b01e01
fix - use anthropic class for clients
2024-04-06 18:19:28 -07:00
Ishaan Jaff
4dfe3f5aee
async ahtropic streaming
2024-04-06 17:54:19 -07:00
Ishaan Jaff
3c10bfb497
async streaming anthropic
2024-04-06 17:53:06 -07:00
Ishaan Jaff
32c3aab34e
feat - make anthropic async
2024-04-06 15:50:13 -07:00
Krish Dholakia
9912a80190
Merge pull request #2855 from Caixiaopig/fix_update_default_claude3_maxtokens
...
Updating the default Anthropic Officlal Claude 3 max_tokens to 4096
2024-04-06 08:39:55 -07:00
Ishaan Jaff
34e3cae0a5
fix - test using specific commit
2024-04-05 16:10:54 -07:00
Krish Dholakia
10951ca2f0
Merge pull request #2856 from lazyhope/anthropic-tools-use-2024-04-04
...
Support latest Anthropic Tools Use (2024-04-04)
2024-04-05 14:31:26 -07:00
Krrish Dholakia
00c0a0c209
fix(vertex_ai_anthropic.py): fix import
2024-04-05 13:38:05 -07:00
Zihao Li
715c11a9da
Add backward compatibility to support xml tool use for bedrock and vertex
2024-04-06 00:34:33 +08:00
Krrish Dholakia
a8d0aace83
fix(vertex_ai_anthropic.py): support streaming, async completion, async streaming for vertex ai anthropic
2024-04-05 09:27:48 -07:00
lazyhope
a7eadf578a
Merge branch 'BerriAI:main' into anthropic-tools-use-2024-04-04
2024-04-05 23:51:03 +08:00
Caixiaopig
aa2a1389da
Updating the default Anthropic Officlal Claude 3 max_tokens to 4096
...
fix bug
2024-04-05 09:45:57 -05:00
Zihao Li
017abaa452
Clean up imports of XML processing functions
2024-04-05 22:36:18 +08:00
Krish Dholakia
5ea9946925
Merge pull request #2665 from BerriAI/litellm_claude_vertex_ai
...
[WIP] feat(vertex_ai_anthropic.py): Add support for claude 3 on vertex ai
2024-04-05 07:06:04 -07:00
Zihao Li
7a034d5942
Refactor tool result submission and tool invoke conversion
2024-04-05 17:11:35 +08:00
Zihao Li
4d0975bf4e
Move tool definitions from system prompt to parameter and refactor tool calling parse
2024-04-05 16:01:40 +08:00
Caixiaopig
7eb9abba2b
Updating the default Anthropic Claude 3 max_tokens to 4096
...
The default value of max_tokens used to be 256. If the client does not set a larger value, the model's output may be truncated, so the default value has been changed to 4096. This value is also the maximum output value described in the official interface.
see: https://docs.anthropic.com/claude/reference/messages_post
2024-04-05 14:44:40 +08:00
Krrish Dholakia
7880dcbdf7
fix(openai.py): support passing prompt as list instead of concat string
2024-04-03 15:23:20 -07:00
Krrish Dholakia
f3fcfe3007
fix(proxy_server.py): return original model response via response headers - /v1/completions
...
to help devs with debugging
2024-04-03 13:05:43 -07:00
Krrish Dholakia
bd7040969b
feat(vertex_ai_anthropic.py): add claude 3 on vertex ai support - working .completions call
...
.completions() call works
2024-04-02 22:07:39 -07:00
yishiyiyuan
8619d29741
🐞 fix: djl vllm support
...
support vllm response format on sagemaker, which only return one choice.
2024-04-03 11:00:51 +08:00
