Add support for GPTQ Marlin (#2052)

Add support for GPTQ Marlin kernels GPTQ Marlin extends the Marlin kernels to support common GPTQ configurations: - bits: 4 or 8 - groupsize: -1, 32, 64, or 128 - desc_act: true/false Using the GPTQ Marlin kernels requires repacking the parameters in the Marlin quantizer format. The kernels were contributed by Neural Magic to VLLM. We vendor them here for convenience.
huggingface · Jun 14, 2024 · 093a27c · 093a27c
1 parent f433f1f
commit 093a27c
Show file tree

Hide file tree

Showing 40 changed files with 4,652 additions and 138 deletions.
diff --git a/Dockerfile b/Dockerfile
@@ -140,9 +140,9 @@ RUN TORCH_CUDA_ARCH_LIST="8.0;8.6+PTX" make build-eetq
 # Build marlin kernels
 FROM kernel-builder as marlin-kernels-builder
 WORKDIR /usr/src
-COPY server/Makefile-marlin Makefile
+COPY server/marlin/ .
 # Build specific version of transformers
-RUN TORCH_CUDA_ARCH_LIST="8.0;8.6+PTX" make build-marlin
+RUN TORCH_CUDA_ARCH_LIST="8.0;8.6+PTX" python setup.py build
 
 # Build Transformers CUDA kernels
 FROM kernel-builder as custom-kernels-builder
@@ -213,7 +213,7 @@ COPY --from=awq-kernels-builder /usr/src/llm-awq/awq/kernels/build/lib.linux-x86
 # Copy build artifacts from eetq kernels builder
 COPY --from=eetq-kernels-builder /usr/src/eetq/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
 # Copy build artifacts from marlin kernels builder
-COPY --from=marlin-kernels-builder /usr/src/marlin/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
+COPY --from=marlin-kernels-builder /usr/src/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
 
 # Copy builds artifacts from vllm builder
 COPY --from=vllm-builder /usr/src/vllm/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages

diff --git a/...tests/models/__snapshots__/test_flash_llama_gptq_marlin/test_flash_llama_gptq_marlin.json b/...tests/models/__snapshots__/test_flash_llama_gptq_marlin/test_flash_llama_gptq_marlin.json
@@ -0,0 +1,84 @@
+{
+  "details": {
+    "best_of_sequences": null,
+    "finish_reason": "length",
+    "generated_tokens": 10,
+    "prefill": [
+      {
+        "id": 2323,
+        "logprob": null,
+        "text": "Test"
+      },
+      {
+        "id": 1715,
+        "logprob": -11.34375,
+        "text": " request"
+      }
+    ],
+    "seed": null,
+    "tokens": [
+      {
+        "id": 198,
+        "logprob": -2.5742188,
+        "special": false,
+        "text": "\n"
+      },
+      {
+        "id": 262,
+        "logprob": -1.6230469,
+        "special": false,
+        "text": "   "
+      },
+      {
+        "id": 3270,
+        "logprob": -2.046875,
+        "special": false,
+        "text": " \"\"\"\n"
+      },
+      {
+        "id": 262,
+        "logprob": -0.015281677,
+        "special": false,
+        "text": "   "
+      },
+      {
+        "id": 422,
+        "logprob": -2.1425781,
+        "special": false,
+        "text": " if"
+      },
+      {
+        "id": 1715,
+        "logprob": -0.9238281,
+        "special": false,
+        "text": " request"
+      },
+      {
+        "id": 13204,
+        "logprob": -0.076660156,
+        "special": false,
+        "text": ".method"
+      },
+      {
+        "id": 624,
+        "logprob": -0.021987915,
+        "special": false,
+        "text": " =="
+      },
+      {
+        "id": 364,
+        "logprob": -0.39208984,
+        "special": false,
+        "text": " '"
+      },
+      {
+        "id": 3019,
+        "logprob": -0.10821533,
+        "special": false,
+        "text": "POST"
+      }
+    ],
+    "top_tokens": null
+  },
+  "generated_text": "\n    \"\"\"\n    if request.method == 'POST"
+}
diff --git a/...s/__snapshots__/test_flash_llama_gptq_marlin/test_flash_llama_gptq_marlin_all_params.json b/...s/__snapshots__/test_flash_llama_gptq_marlin/test_flash_llama_gptq_marlin_all_params.json
@@ -0,0 +1,84 @@
+{
+  "details": {
+    "best_of_sequences": null,
+    "finish_reason": "length",
+    "generated_tokens": 10,
+    "prefill": [
+      {
+        "id": 2323,
+        "logprob": null,
+        "text": "Test"
+      },
+      {
+        "id": 1715,
+        "logprob": -11.34375,
+        "text": " request"
+      }
+    ],
+    "seed": 0,
+    "tokens": [
+      {
+        "id": 13,
+        "logprob": -2.2539062,
+        "special": false,
+        "text": "."
+      },
+      {
+        "id": 578,
+        "logprob": -0.15563965,
+        "special": false,
+        "text": " The"
+      },
+      {
+        "id": 3622,
+        "logprob": -0.8203125,
+        "special": false,
+        "text": " server"
+      },
+      {
+        "id": 706,
+        "logprob": 0.0,
+        "special": false,
+        "text": " has"
+      },
+      {
+        "id": 539,
+        "logprob": 0.0,
+        "special": false,
+        "text": " not"
+      },
+      {
+        "id": 3686,
+        "logprob": 0.0,
+        "special": false,
+        "text": " yet"
+      },
+      {
+        "id": 3288,
+        "logprob": 0.0,
+        "special": false,
+        "text": " sent"
+      },
+      {
+        "id": 904,
+        "logprob": 0.0,
+        "special": false,
+        "text": " any"
+      },
+      {
+        "id": 828,
+        "logprob": 0.0,
+        "special": false,
+        "text": " data"
+      },
+      {
+        "id": 382,
+        "logprob": -1.5517578,
+        "special": false,
+        "text": ".\n\n"
+      }
+    ],
+    "top_tokens": null
+  },
+  "generated_text": "Test request. The server has not yet sent any data.\n\n"
+}