bitsandbytes-cpu-fork 0.50.2.dev1__tar.gz → 0.50.2.dev5__tar.gz

{bitsandbytes_cpu_fork-0.50.2.dev1/bitsandbytes_cpu_fork.egg-info → bitsandbytes_cpu_fork-0.50.2.dev5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: bitsandbytes-cpu-fork
-Version: 0.50.2.dev1
+Version: 0.50.2.dev5
 Summary: bitsandbytes with a pure-CPU training backend (AVX2 / NEON): 8-bit optimizers, fused GDN, gemm_8bit, disk balancer. No GPU required.
 License-Expression: MIT
 Project-URL: repository, https://github.com/Ganyvnanoaa09-0906/bitsandbytes-CPU
@@ -49,6 +49,13 @@ pip install torch --index-url https://download.pytorch.org/whl/cpu
 pip install bitsandbytes-cpu-fork
 ```
 
+**Linux wheels.** This release ships two Linux x86_64 wheels: a native one
+(`manylinux_2_39`) with the OpenMP-parallel kernels, and a portable one
+(`manylinux_2_17`, single-threaded) for older distributions. pip picks the best one
+for the host automatically. If neither matches your glibc, install from the sdist
+(it compiles `csrc` with the local toolchain via `build_linux.sh`).
+
+
 ### Windows
 
 A wheel is published with the CPU library already built in, so no compiler and no
@@ -182,7 +189,7 @@ bitsandbytes has the following minimum requirements for all platforms:
       <td>🟥 AMD GPU <br><code>cuda</code></td>
       <td>
         CDNA: gfx908, gfx90a, gfx942, gfx950, gfx1250<br>
-        RDNA: gfx103X, gfx110X, gfx115X, gfx120X
+        RDNA: gfx101X, gfx103X, gfx110X, gfx115X, gfx120X
       </td>
       <td>✅</td>
       <td>✅</td>
@@ -247,7 +254,8 @@ bitsandbytes has the following minimum requirements for all platforms:
       <td></td>
       <td>🟥 AMD GPU <br><code>cuda</code></td>
       <td>
-        RDNA: gfx103X, gfx110X, gfx115X, gfx120X
+        CDNA: gfx908, gfx90a<br>
+        RDNA: gfx101X, gfx103X, gfx110X, gfx115X, gfx120X
       </td>
       <td>✅</td>
       <td>✅</td>
@@ -328,7 +336,7 @@ code and the Transformers / PEFT / Diffusers integrations keep working as-is:
 
 ```python
 import bitsandbytes as bnb          # the fork, not upstream
-print(bnb.__version__)              # 0.50.2.dev1
+print(bnb.__version__)              # 0.50.2.dev5
 ```
 
 Building from source instead (any platform) is described in
@@ -373,7 +381,7 @@ build_manual\build_manual.bat amd   # (in VS x64 Native Tools prompt)
 
 > `.gitattributes` 强制 C/C++ 源码与构建脚本用 LF 换行，clone 到 Windows/Linux 均可
 > 直接编译（CRLF 会让 MSVC 吞行、Linux `g++` 报 `bad interpreter`）。
-> `disk_balancer`（`--flash`）跨平台：Windows 用 ctypes，Linux/WSL 用 psutil。
+> `disk_balancer`（`--flash`）跨平台：Windows 用 ctypes；Linux 用 `/proc/mounts` + psutil + `/sys/block/*/queue/rotational`；**WSL 已支持**——缓存只落在 WSL 自己的 ext4（默认 `~/.cache/bnb_flash_cache`），9P 宿主盘 `/mnt/[盘符]`、网络盘、只读挂载、tmpfs 一律拒绝（显式 `--flash_paths` 指过去会直接报错）。零拷贝写盘队列有在飞字节上限（`max_inflight_bytes`），卸载不会把源张量内存偷偷留住。
 
 ### Prebuilt wheels (optional)