seedpack 0.2.1__tar.gz → 0.2.2__tar.gz

{seedpack-0.2.1/seedpack.egg-info → seedpack-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: seedpack
-Version: 0.2.1
+Version: 0.2.2
 Summary: Lossless GPU archiver for GGUF model repositories: the BF16 model and all its quants in one archive about the size of a single compressed BF16 file
 Author: NeuroSquad
 License-Expression: Apache-2.0

{seedpack-0.2.1 → seedpack-0.2.2}/seedpack/__init__.py

@@ -3,4 +3,4 @@
 The quants of a repository are re-derived on the GPU from its most precise model (the seed) with
 a bit-exact port of llama.cpp's quantisers; only what differs is stored. Command line: `seedpack`.
 """
-__version__ = "0.2.1"
+__version__ = "0.2.2"

{seedpack-0.2.1 → seedpack-0.2.2}/seedpack/repo3.py

@@ -1712,8 +1712,16 @@ class M28RepoGPU:
             _release()
             return win
 
+        fetched = {}                          # the coded windows read ahead (two: the drive never waits)
+
+        def fetch(w):
+            if w < len(wins) and w not in fetched:
+                fetched[w] = bread.submit(self._window_fetch, rt, order[wins[w][0]:wins[w][1]])
+
         def ahead(w):
-            return gpu.submit(prepare, w, bread.submit(self._window_fetch, rt, order[wins[w][0]:wins[w][1]]))
+            fetch(w)
+            fetch(w + 1)
+            return gpu.submit(prepare, w, fetched.pop(w))
 
         wb = [sum(self.by_name[n].nbytes for n in order[a:z]) for a, z in wins]
         limit = 2 * self._window_bytes()

{seedpack-0.2.1 → seedpack-0.2.2/seedpack.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: seedpack
-Version: 0.2.1
+Version: 0.2.2
 Summary: Lossless GPU archiver for GGUF model repositories: the BF16 model and all its quants in one archive about the size of a single compressed BF16 file
 Author: NeuroSquad
 License-Expression: Apache-2.0