inference: add AGPLv3 license headers

All eight files introduced on this branch now carry the SPDX AGPLv3
header used by the MoE kernels. flex_paged_attention.py keeps its
BSD 3-Clause attribution to attention-gym alongside the new header.

tests/flex_fastlm_bench.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+
 """Batched steady-state throughput bench through ``FastLanguageModel`` +
 ``UNSLOTH_FAST_INFERENCE=1``. First ``generate`` call primes CUDA graphs;
 subsequent calls report steady state. Compare against April CLI-only

tests/flex_fastlm_smoke.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+
 """Smoke-test the ``UNSLOTH_FAST_INFERENCE=1`` path through
 ``FastLanguageModel.from_pretrained``.
 

unsloth/inference/__init__.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+
 """Flex-attention inference engines.
 
 ``UNSLOTH_FAST_INFERENCE=1`` routes ``FastLanguageModel.from_pretrained``

unsloth/inference/flex_engine.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+
 """FlexEngine: vLLM-compatible LLM surface for the flex inference backends.
 
 When ``UNSLOTH_FAST_INFERENCE=1`` is set, :func:`load_flex` wraps the HF model

unsloth/inference/flex_gemma4.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+
 """Gemma-4-E2B-it inference with flex_attention + paged KV cache + CUDA graphs.
 
 Extends the Qwen3/Llama-3.2 engine in `qwen3_flex_inference.py` to a third

unsloth/inference/flex_paged_attention.py

@@ -1,7 +1,9 @@
-# Adapted from attention-gym
-# Original source: https://github.com/pytorch-labs/attention-gym
-# License: BSD 3-Clause (see THIRD_PARTY_LICENSES.md)
-# Copyright (c) 2023, Driss Guessous
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+#
+# Adapted from attention-gym (https://github.com/pytorch-labs/attention-gym)
+# Copyright (c) 2023, Driss Guessous, licensed under BSD 3-Clause
+# (see THIRD_PARTY_LICENSES.md).
 
 # the original implementation has some bugs and has some feature that lives outside of the PageTable class
 

unsloth/inference/flex_qwen3_llama.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+
 """Llama / Qwen3 inference with flex_attention + paged KV cache + CUDA graphs.
 
 The transformers continuous-batching path tops out at ~10% of vLLM on this

unsloth/inference/vllm_shim.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: GNU Affero General Public License v3.0
+# Copyright 2023-present the Unsloth team. All rights reserved.
+
 """vLLM-API surface for the flex inference backend.
 
 `FlexEngine.generate` / `.chat` return :class:`RequestOutput` objects with the