Basic flashinfer 0.2 support

This change does not use any of the new features yet, but makes
some small compatibility changes.

flake.nix

@@ -5,7 +5,7 @@
       inputs.nixpkgs.follows = "tgi-nix/nixpkgs";
     };
     nix-filter.url = "github:numtide/nix-filter";
-    tgi-nix.url = "github:huggingface/text-generation-inference-nix";
+    tgi-nix.url = "github:huggingface/text-generation-inference-nix/flashinfer-v0.2";
     nixpkgs.follows = "tgi-nix/nixpkgs";
     flake-utils.url = "github:numtide/flake-utils";
     rust-overlay = {

server/Makefile-flashinfer

@@ -1,2 +1,4 @@
 install-flashinfer:
-	pip install flashinfer==0.1.6 -i https://flashinfer.ai/whl/cu124/torch2.4
+	# Avoid: Could not find a version that satisfies the requirement fsspec (from torch).
+	pip install fsspec
+	pip install flashinfer==0.2.0 -i https://flashinfer.ai/whl/cu124/torch2.4

server/text_generation_server/layers/attention/flashinfer.py

@@ -93,7 +93,7 @@ def use_prefill_with_paged_kv_state(
             head_dim=head_size,
             q_data_type=dtype,
             page_size=page_size,
-            window_left=window_left,
+            window_left=-1 if window_left is None else window_left,
         )
         yield
     finally:
@@ -139,7 +139,7 @@ def use_prefill_state(
             num_kv_heads=num_kv_heads,
             head_dim=head_size,
             q_data_type=dtype,
-            window_left=window_left,
+            window_left=-1 if window_left is None else window_left,
         )
         yield
     finally:
@@ -243,7 +243,7 @@ def use_decode_state(
             page_size=page_size,
             data_type=kv_cache_dtype,
             q_data_type=dtype,
-            window_left=window_left,
+            window_left=-1 if window_left is None else window_left,
         )
         yield
     finally: