diff --git a/pkgs/development/python-modules/openllm-client/default.nix b/pkgs/development/python-modules/openllm-client/default.nix
index 2dd395bab677..ce77953f12df 100644
--- a/pkgs/development/python-modules/openllm-client/default.nix
+++ b/pkgs/development/python-modules/openllm-client/default.nix
@@ -5,11 +5,11 @@
 , hatch-fancy-pypi-readme
 , hatch-vcs
 , hatchling
-, attrs
-, cattrs
+, anyio
+, distro
 , httpx
+, httpx-auth
 , openllm-core
-, orjson
 , soundfile
 , transformers
 }:
@@ -30,16 +30,19 @@ buildPythonPackage rec {
   ];
 
   propagatedBuildInputs = [
-    attrs
-    cattrs
+    anyio
+    distro
     httpx
-    orjson
+    openllm-core
   ];
 
   passthru.optional-dependencies = {
     grpc = [
       bentoml
     ] ++ bentoml.optional-dependencies.grpc;
+    auth = [
+      httpx-auth
+    ];
     agents = [
       transformers
       # diffusers
diff --git a/pkgs/development/python-modules/openllm-core/default.nix b/pkgs/development/python-modules/openllm-core/default.nix
index 75b755740d04..ab39bf6c11b9 100644
--- a/pkgs/development/python-modules/openllm-core/default.nix
+++ b/pkgs/development/python-modules/openllm-core/default.nix
@@ -4,10 +4,12 @@
 , pythonOlder
 , accelerate
 , attrs
-, bentoml
 , bitsandbytes
+, bentoml
 , cattrs
+, click-option-group
 , datasets
+, deepmerge
 , hatch-fancy-pypi-readme
 , hatch-vcs
 , hatchling
@@ -15,14 +17,13 @@
 , mypy-extensions
 , orjson
 , peft
-, ray
 , transformers
 , typing-extensions
 }:
 
 buildPythonPackage rec {
   pname = "openllm-core";
-  version = "0.3.9";
+  version = "0.4.22";
   pyproject = true;
 
   disabled = pythonOlder "3.8";
@@ -31,7 +32,7 @@ buildPythonPackage rec {
     owner = "bentoml";
     repo = "OpenLLM";
     rev = "refs/tags/v${version}";
-    hash = "sha256-M/ckvaHTdKFg7xfUgFxu7pRBrS6TGw0m2U3L88b2DKU=";
+    hash = "sha256-Hgwc4rneY0d7KZHuBIWRpndLksts5DTvaYuwZOO4sdI=";
   };
 
   sourceRoot = "source/openllm-core";
@@ -44,8 +45,10 @@ buildPythonPackage rec {
 
   propagatedBuildInputs = [
     attrs
-    bentoml
     cattrs
+    # not listed in pyproject.toml, but required at runtime
+    click-option-group
+    deepmerge
     inflection
     mypy-extensions
     orjson
@@ -54,9 +57,11 @@ buildPythonPackage rec {
 
   passthru.optional-dependencies = {
     vllm = [
-      ray
       # vllm
     ];
+    bentoml = [
+      bentoml
+    ];
     fine-tune = [
       accelerate
       bitsandbytes
@@ -67,7 +72,7 @@ buildPythonPackage rec {
     ] ++ transformers.optional-dependencies.torch
       ++ transformers.optional-dependencies.tokenizers
       ++ transformers.optional-dependencies.accelerate;
-    full = with passthru.optional-dependencies; ( vllm ++ fine-tune );
+    full = with passthru.optional-dependencies; ( vllm ++ bentoml ++ fine-tune );
   };
 
   # there is no tests
diff --git a/pkgs/development/python-modules/openllm/default.nix b/pkgs/development/python-modules/openllm/default.nix
index b9f3d2b6fa3b..b1cad320a11d 100644
--- a/pkgs/development/python-modules/openllm/default.nix
+++ b/pkgs/development/python-modules/openllm/default.nix
@@ -9,20 +9,18 @@
 , accelerate
 , bentoml
 , bitsandbytes
+, build
 , click
+, ctranslate2
 , datasets
 , docker
 , einops
-, fairscale
-, flax
 , ghapi
+, huggingface-hub
 , hypothesis
 , ipython
-, jax
-, jaxlib
 , jupyter
 , jupytext
-, keras
 , nbformat
 , notebook
 , openai
@@ -34,14 +32,12 @@
 , pytest-randomly
 , pytest-rerunfailures
 , pytest-xdist
-, ray
 , safetensors
 , scipy
 , sentencepiece
 , soundfile
 , syrupy
 , tabulate
-, tensorflow
 , tiktoken
 , transformers
 , openai-triton
@@ -73,12 +69,16 @@ buildPythonPackage rec {
     accelerate
     bentoml
     bitsandbytes
+    build
     click
+    einops
     ghapi
     openllm-client
     openllm-core
     optimum
     safetensors
+    scipy
+    sentencepiece
     tabulate
     transformers
   ] ++ bentoml.optional-dependencies.io
@@ -92,62 +92,43 @@ buildPythonPackage rec {
       soundfile
       transformers
     ] ++ transformers.optional-dependencies.agents;
+    awq = [
+      # autoawq
+    ];
     baichuan = [
       # cpm-kernels
-      sentencepiece
     ];
     chatglm = [
       # cpm-kernels
-      sentencepiece
+    ];
+    ctranslate = [
+      ctranslate2
     ];
     falcon = [
-      einops
       xformers
     ];
     fine-tune = [
-      accelerate
-      bitsandbytes
       datasets
+      huggingface-hub
       peft
       # trl
     ];
-    flan-t5 = [
-      flax
-      jax
-      jaxlib
-      keras
-      tensorflow
-    ];
     ggml = [
       # ctransformers
     ];
     gptq = [
       # auto-gptq
-      optimum
     ]; # ++ autogptq.optional-dependencies.triton;
     grpc = [
-      openllm-client
-    ] ++ openllm-client.optional-dependencies.grpc;
-    llama = [
-      fairscale
-      sentencepiece
-      scipy
-    ];
+      bentoml
+    ] ++ bentoml.optional-dependencies.grpc;
     mpt = [
-      einops
       openai-triton
     ];
     openai = [
       openai
       tiktoken
-    ] ++ openai.optional-dependencies.embeddings;
-    opt = [
-      flax
-      jax
-      jaxlib
-      keras
-      tensorflow
-    ];
+    ] ++ openai.optional-dependencies.datalib;
     playground = [
       ipython
       jupyter
@@ -159,11 +140,10 @@ buildPythonPackage rec {
       bitsandbytes
     ];
     vllm = [
-      ray
       # vllm
     ];
     full = with passthru.optional-dependencies; (
-      agents ++ baichuan ++ chatglm ++ falcon ++ fine-tune ++ flan-t5 ++ ggml ++ gptq ++ llama ++ mpt ++ openai ++ opt ++ playground ++ starcoder ++ vllm
+      agents ++ awq ++ baichuan ++ chatglm ++ ctranslate ++ falcon ++ fine-tune ++ ggml ++ gptq ++ mpt ++ openai ++ playground ++ starcoder ++ vllm
     );
     all = passthru.optional-dependencies.full;
   };
@@ -187,12 +167,15 @@ buildPythonPackage rec {
     export CI=1
   '';
 
+  disabledTestPaths = [
+    # require network access
+    "tests/models"
+  ];
+
   disabledTests = [
-    # these tests access to huggingface.co
-    "test_opt_125m"
-    "test_opt_125m"
-    "test_flan_t5"
-    "test_flan_t5"
+    # incompatible with recent TypedDict
+    # https://github.com/bentoml/OpenLLM/blob/f3fd32d596253ae34c68e2e9655f19f40e05f666/openllm-python/tests/configuration_test.py#L18-L21
+    "test_missing_default"
   ];
 
   pythonImportsCheck = [ "openllm" ];