Fix Nemotron cache leak upstream (#1819)

d9ed943034b453c87ff597f0545d913b408882b1 · 2026-03-30 17:53:21 +0100 · rltakashige

## Motivation
Nemotron Cascade and Nano failing at long decodes.

## Changes

Fixed upstream, just change pyproject and uv lock here.


## Test Plan
### Automated Testing
Tested with a reproduce script upstream

Files touched

M pyproject.toml
M uv.lock

Diff

commit d9ed943034b453c87ff597f0545d913b408882b1
Author: rltakashige <rl.takashige@gmail.com>
Date:   Mon Mar 30 17:53:21 2026 +0100

    Fix Nemotron cache leak upstream (#1819)
    
    ## Motivation
    Nemotron Cascade and Nano failing at long decodes.
    
    ## Changes
    
    Fixed upstream, just change pyproject and uv lock here.
    
    
    ## Test Plan
    ### Automated Testing
    Tested with a reproduce script upstream
---
 pyproject.toml | 2 +-
 uv.lock        | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/pyproject.toml b/pyproject.toml
index 9895e0c7..ecb66133 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -63,7 +63,7 @@ members = ["rust/exo_pyo3_bindings", "bench"]
 [tool.uv.sources]
 exo_pyo3_bindings = { workspace = true }
 mlx = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git", branch = "address-rdma-gpu-locks", marker = "sys_platform == 'darwin'" }
-mlx-lm = { git = "https://github.com/rltakashige/mlx-lm", branch = "leo/fix-deepseek-v32-indexer" }
+mlx-lm = { git = "https://github.com/rltakashige/mlx-lm", branch = "leo/fix-arrayscache-leak" }
 # Uncomment to use local mlx/mlx-lm development versions:
 # mlx = { path = "/Users/Shared/mlx", editable=true }
 # mlx-lm = { path = "/Users/Shared/mlx-lm", editable=true }
diff --git a/uv.lock b/uv.lock
index 7ddd1486..133393ef 100644
--- a/uv.lock
+++ b/uv.lock
@@ -599,7 +599,7 @@ requires-dist = [
     { name = "mflux", specifier = "==0.17.2" },
     { name = "mlx", marker = "sys_platform == 'darwin'", git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git?branch=address-rdma-gpu-locks" },
     { name = "mlx", extras = ["cpu"], marker = "sys_platform == 'linux'", specifier = "==0.30.6" },
-    { name = "mlx-lm", git = "https://github.com/rltakashige/mlx-lm?branch=leo%2Ffix-deepseek-v32-indexer" },
+    { name = "mlx-lm", git = "https://github.com/rltakashige/mlx-lm?branch=leo%2Ffix-arrayscache-leak" },
     { name = "mlx-vlm", specifier = ">=0.3.11" },
     { name = "msgspec", specifier = ">=0.19.0" },
     { name = "openai-harmony", specifier = ">=0.0.8" },
@@ -1528,7 +1528,7 @@ wheels = [
 [[package]]
 name = "mlx-lm"
 version = "0.31.2"
-source = { git = "https://github.com/rltakashige/mlx-lm?branch=leo%2Ffix-deepseek-v32-indexer#d388ff77858fec3b5d2e3b1d9502a7e2878b8109" }
+source = { git = "https://github.com/rltakashige/mlx-lm?branch=leo%2Ffix-arrayscache-leak#d36e9b661e55a5fc0f77fb6f17ea643aa2dc87aa" }
 dependencies = [
     { name = "jinja2", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "mlx", version = "0.31.2.dev20260324+e5e64331", source = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git?branch=address-rdma-gpu-locks#e5e64331830d9b04ae9082b843073f9c1fa7705e" }, marker = "sys_platform == 'darwin'" },

← c6815bfd Only update KV prefix cache on a good cache hit (#1817) · back to Exo · Support PDFs in dashboard (#1822) 4688adb5 →