diff --git a/Cargo.lock b/Cargo.lock index 6088d7b11f6..de54c536545 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -3393,7 +3393,7 @@ dependencies = [ [[package]] name = "text-generation-benchmark" -version = "2.0.0" +version = "2.0.1" dependencies = [ "average", "clap", @@ -3414,7 +3414,7 @@ dependencies = [ [[package]] name = "text-generation-client" -version = "2.0.0" +version = "2.0.1" dependencies = [ "futures", "grpc-metadata", @@ -3430,7 +3430,7 @@ dependencies = [ [[package]] name = "text-generation-launcher" -version = "2.0.0" +version = "2.0.1" dependencies = [ "clap", "ctrlc", @@ -3448,7 +3448,7 @@ dependencies = [ [[package]] name = "text-generation-router" -version = "2.0.0" +version = "2.0.1" dependencies = [ "async-stream", "axum", diff --git a/Cargo.toml b/Cargo.toml index fffe4ef7091..593fd950c97 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -9,7 +9,7 @@ members = [ resolver = "2" [workspace.package] -version = "2.0.0" +version = "2.0.1" edition = "2021" authors = ["Olivier Dehaene"] homepage = "https://github.com/huggingface/text-generation-inference" diff --git a/docs/openapi.json b/docs/openapi.json index 3093e7e2101..2a387c2f098 100644 --- a/docs/openapi.json +++ b/docs/openapi.json @@ -10,7 +10,7 @@ "name": "Apache 2.0", "url": "https://www.apache.org/licenses/LICENSE-2.0" }, - "version": "2.0.0" + "version": "2.0.1" }, "paths": { "/": { diff --git a/integration-tests/models/__snapshots__/test_chat_llama/test_flash_llama_simple.json b/integration-tests/models/__snapshots__/test_chat_llama/test_flash_llama_simple.json index 0ff874f1c8e..4cb548d2139 100644 --- a/integration-tests/models/__snapshots__/test_chat_llama/test_flash_llama_simple.json +++ b/integration-tests/models/__snapshots__/test_chat_llama/test_flash_llama_simple.json @@ -17,7 +17,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native", + "system_fingerprint": "2.0.1-native", "usage": { "completion_tokens": 100, "prompt_tokens": 60, diff --git a/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts.json b/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts.json index 3fe0a4826d5..99c33cf7582 100644 --- a/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts.json +++ b/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts.json @@ -29,7 +29,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native", + "system_fingerprint": "2.0.1-native", "usage": { "completion_tokens": 36, "prompt_tokens": 8, diff --git a/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts_stream.json b/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts_stream.json index 702d48f4106..d87071cfac5 100644 --- a/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts_stream.json +++ b/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_many_prompts_stream.json @@ -12,7 +12,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -27,7 +27,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -42,7 +42,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -57,7 +57,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -72,7 +72,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -87,7 +87,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -102,7 +102,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -117,7 +117,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -132,7 +132,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -147,7 +147,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -162,7 +162,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -177,7 +177,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -192,7 +192,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -207,7 +207,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -222,7 +222,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -237,7 +237,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -252,7 +252,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -267,7 +267,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -282,7 +282,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -297,7 +297,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -312,7 +312,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -327,7 +327,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -342,7 +342,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -357,7 +357,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -372,7 +372,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -387,7 +387,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -402,7 +402,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -417,7 +417,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -432,7 +432,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -447,7 +447,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -462,7 +462,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -477,7 +477,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -492,7 +492,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -507,7 +507,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -522,7 +522,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -537,7 +537,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -552,7 +552,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -567,7 +567,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -582,7 +582,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" }, { "choices": [ @@ -597,6 +597,6 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" } ] diff --git a/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_single_prompt.json b/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_single_prompt.json index 8b001a09412..5aed493563e 100644 --- a/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_single_prompt.json +++ b/integration-tests/models/__snapshots__/test_completion_prompts/test_flash_llama_completion_single_prompt.json @@ -11,7 +11,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native", + "system_fingerprint": "2.0.1-native", "usage": { "completion_tokens": 5, "prompt_tokens": 6, diff --git a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools.json b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools.json index 45f8ca995e8..a4c34a10e7f 100644 --- a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools.json +++ b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools.json @@ -30,7 +30,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native", + "system_fingerprint": "2.0.1-native", "usage": { "completion_tokens": 37, "prompt_tokens": 524, diff --git a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_auto.json b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_auto.json index e0ed0947fa8..04bcdc4e773 100644 --- a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_auto.json +++ b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_auto.json @@ -30,7 +30,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native", + "system_fingerprint": "2.0.1-native", "usage": { "completion_tokens": 37, "prompt_tokens": 524, diff --git a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_choice.json b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_choice.json index b70c2d6fffb..603c90afa61 100644 --- a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_choice.json +++ b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_choice.json @@ -30,7 +30,7 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native", + "system_fingerprint": "2.0.1-native", "usage": { "completion_tokens": 48, "prompt_tokens": 320, diff --git a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_stream.json b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_stream.json index 6787b39b1d7..f72a5d383c6 100644 --- a/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_stream.json +++ b/integration-tests/models/__snapshots__/test_tools_llama/test_flash_llama_grammar_tools_stream.json @@ -23,5 +23,5 @@ "id": "", "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "object": "text_completion", - "system_fingerprint": "2.0.0-native" + "system_fingerprint": "2.0.1-native" } diff --git a/integration-tests/pyproject.toml b/integration-tests/pyproject.toml index 6923ff23844..88e9761a099 100644 --- a/integration-tests/pyproject.toml +++ b/integration-tests/pyproject.toml @@ -1,6 +1,6 @@ [tool.poetry] name = "text-generation-integration-tests" -version = "2.0.0" +version = "2.0.1" description = "Text Generation Inference integration tests" authors = ["Nicolas Patry "] diff --git a/server/pyproject.toml b/server/pyproject.toml index ccfccdb2bf6..23dc5eee872 100644 --- a/server/pyproject.toml +++ b/server/pyproject.toml @@ -1,6 +1,6 @@ [tool.poetry] name = "text-generation-server" -version = "2.0.0" +version = "2.0.1" description = "Text Generation Inference Python gRPC Server" authors = ["Olivier Dehaene "]