Fix cache offset casting with low precision policies

lib/bumblebee/audio/whisper.ex

@@ -398,7 +398,7 @@ defmodule Bumblebee.Audio.Whisper do
           offset =
             case offset do
               %Axon.None{} -> 0
-              offset -> Nx.as_type(offset, {:s, 64})
+              offset -> offset
             end
 
           input_sequence_length = Nx.axis_size(input_embeddings, 1)

lib/bumblebee/layers.ex

@@ -120,7 +120,7 @@ defmodule Bumblebee.Layers do
             bias
 
           offset ->
-            mask_shift = Nx.as_type(offset, {:s, 64})
+            mask_shift = offset
             query_length = Nx.axis_size(query, 1)
             Nx.slice_along_axis(bias, mask_shift, query_length, axis: 2)
         end

lib/bumblebee/layers/decoder.ex

@@ -82,9 +82,9 @@ defmodule Bumblebee.Layers.Decoder do
       |> List.duplicate(decoder_num_blocks)
       |> List.to_tuple()
 
-    offset = Nx.tensor(0.0)
+    offset = Nx.tensor(0)
 
-    attention_mask = Nx.broadcast(0.0, {batch_size, max_length})
+    attention_mask = Nx.broadcast(0, {batch_size, max_length})
 
     %{blocks: blocks, offset: offset, attention_mask: attention_mask}
   end
@@ -170,7 +170,7 @@ defmodule Bumblebee.Layers.Decoder do
 
   defnp append_attention_cache(key, value, attention_cache, offset, _opts \\ []) do
     %{key: cached_key, value: cached_value} = attention_cache
-    indices = [0, Nx.as_type(offset, {:s, 64}), 0, 0]
+    indices = [0, offset, 0, 0]
     key = Nx.put_slice(cached_key, indices, key)
     value = Nx.put_slice(cached_value, indices, value)
     updated_cache = %{key: key, value: value}
@@ -276,7 +276,7 @@ defmodule Bumblebee.Layers.Decoder do
                 causal_mask
 
               offset ->
-                mask_shift = Nx.as_type(offset, {:s, 64})
+                mask_shift = offset
                 query_length = Nx.axis_size(query, 1)
                 Nx.slice_along_axis(causal_mask, mask_shift, query_length, axis: 2)
             end

mix.exs

@@ -30,7 +30,8 @@ defmodule Bumblebee.MixProject do
 
   defp deps do
     [
-      {:axon, "~> 0.6.0", axon_opts()},
+      # {:axon, "~> 0.6.0", axon_opts()},
+      {:axon, github: "elixir-nx/axon", override: true},
       {:tokenizers, "~> 0.4"},
       # {:nx, "~> 0.6.2"},
       # {:exla, ">= 0.0.0", only: [:dev, :test]},

mix.lock

@@ -1,5 +1,5 @@
 %{
-  "axon": {:hex, :axon, "0.6.0", "fd7560079581e4cedebaf0cd5f741d6ac3516d06f204ebaf1283b1093bf66ff6", [:mix], [{:kino, "~> 0.7", [hex: :kino, repo: "hexpm", optional: true]}, {:kino_vega_lite, "~> 0.1.7", [hex: :kino_vega_lite, repo: "hexpm", optional: true]}, {:nx, "~> 0.6.0", [hex: :nx, repo: "hexpm", optional: false]}, {:polaris, "~> 0.1", [hex: :polaris, repo: "hexpm", optional: false]}, {:table_rex, "~> 3.1.1", [hex: :table_rex, repo: "hexpm", optional: true]}], "hexpm", "204e7aeb50d231a30b25456adf17bfbaae33fe7c085e03793357ac3bf62fd853"},
+  "axon": {:git, "https://github.com/elixir-nx/axon.git", "67b48c7a43438f5eec2a35311572565cafe889d7", []},
   "bypass": {:hex, :bypass, "2.1.0", "909782781bf8e20ee86a9cabde36b259d44af8b9f38756173e8f5e2e1fabb9b1", [:mix], [{:plug, "~> 1.7", [hex: :plug, repo: "hexpm", optional: false]}, {:plug_cowboy, "~> 2.0", [hex: :plug_cowboy, repo: "hexpm", optional: false]}, {:ranch, "~> 1.3", [hex: :ranch, repo: "hexpm", optional: false]}], "hexpm", "d9b5df8fa5b7a6efa08384e9bbecfe4ce61c77d28a4282f79e02f1ef78d96b80"},
   "castore": {:hex, :castore, "1.0.4", "ff4d0fb2e6411c0479b1d965a814ea6d00e51eb2f58697446e9c41a97d940b28", [:mix], [], "hexpm", "9418c1b8144e11656f0be99943db4caf04612e3eaecefb5dae9a2a87565584f8"},
   "cc_precompiler": {:hex, :cc_precompiler, "0.1.8", "933a5f4da3b19ee56539a076076ce4d7716d64efc8db46fd066996a7e46e2bfd", [:mix], [{:elixir_make, "~> 0.7.3", [hex: :elixir_make, repo: "hexpm", optional: false]}], "hexpm", "176bdf4366956e456bf761b54ad70bc4103d0269ca9558fd7cee93d1b3f116db"},