update version and requirements-rocm.txt

CMakeLists.txt

@@ -191,7 +191,7 @@ set(VLLM_EXT_SRC
   "csrc/attention/attention_kernels_opt.cu"
   "csrc/opt/layernorm_kernels_opt.cu"
   "csrc/quantization/squeezellm/quant_cuda_kernel.cu"
-  "csrc/quantization/gptq/q_gemm.cu"
+  # "csrc/quantization/gptq/q_gemm.cu"
   "csrc/quantization/compressed_tensors/int8_quant_kernels.cu"
   # "csrc/quantization/fp8/common.cu"
   "csrc/cuda_utils_kernels.cu"

csrc/ops.h

@@ -215,12 +215,12 @@ void dynamic_scaled_int8_quant(torch::Tensor& out, torch::Tensor const& input,
 void squeezellm_gemm(torch::Tensor vec, torch::Tensor mat, torch::Tensor mul,
                      torch::Tensor lookup_table);
 
-torch::Tensor gptq_gemm(torch::Tensor a, torch::Tensor b_q_weight,
-                        torch::Tensor b_gptq_qzeros,
-                        torch::Tensor b_gptq_scales, torch::Tensor b_g_idx,
-                        bool use_exllama, int64_t bit);
+// torch::Tensor gptq_gemm(torch::Tensor a, torch::Tensor b_q_weight,
+//                         torch::Tensor b_gptq_qzeros,
+//                         torch::Tensor b_gptq_scales, torch::Tensor b_g_idx,
+//                         bool use_exllama, int64_t bit);
 
-void gptq_shuffle(torch::Tensor q_weight, torch::Tensor q_perm, int64_t bit);
+// void gptq_shuffle(torch::Tensor q_weight, torch::Tensor q_perm, int64_t bit);
 
 // void static_scaled_fp8_quant(torch::Tensor& out, torch::Tensor const& input,
 //                              torch::Tensor const& scale);

csrc/torch_bindings.cpp

@@ -295,12 +295,12 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
 #endif
 
   // Quantized GEMM for GPTQ.
-  ops.def("gptq_gemm", &gptq_gemm);
-  ops.impl("gptq_gemm", torch::kCUDA, &gptq_gemm);
+//   ops.def("gptq_gemm", &gptq_gemm);
+//   ops.impl("gptq_gemm", torch::kCUDA, &gptq_gemm);
 
   // Post processing for GPTQ.
-  ops.def("gptq_shuffle(Tensor! q_weight, Tensor q_perm, int bit) -> ()");
-  ops.impl("gptq_shuffle", torch::kCUDA, &gptq_shuffle);
+//   ops.def("gptq_shuffle(Tensor! q_weight, Tensor q_perm, int bit) -> ()");
+//   ops.impl("gptq_shuffle", torch::kCUDA, &gptq_shuffle);
 
   // Quantized GEMM for SqueezeLLM.
   ops.def(

requirements-rocm.txt

@@ -9,3 +9,9 @@ ray >= 2.10.0
 peft
 pytest-asyncio
 tensorizer>=2.9.0
+
+torch == 2.1.0
+triton == 2.1.0
+flash_attn == 2.0.4
+xformers == 0.0.25
+lmslim == 0.1.0
\ No newline at end of file

setup.py

@@ -375,13 +375,21 @@ def get_sha(root: Union[str, Path]) -> str:
 def get_version_add(sha: Optional[str] = None) -> str:
     vllm_root = os.path.dirname(os.path.abspath(__file__))
     add_version_path = os.path.join(os.path.join(vllm_root, "vllm"), "version.py")
+    major, minor, *res = torch.__version__.split('.')
     if add_git_version:
         if sha != 'Unknown':
             if sha is None:
                 sha = get_sha(vllm_root)
-            version = 'das.opt1' + sha[:7]
+            if (major, minor) == ('2', '1'):
+                version = 'das.opt1.' + sha[:7]
+            if (major, minor) == ('2', '3'):
+                version = 'das.opt2.' + sha[:7]
     else:
+        if (major, minor) == ('2', '1'):
             version = 'das.opt1'
+        if (major, minor) == ('2', '3'):
+            version = 'das.opt2'
+
 
     # dtk version
     if os.getenv("ROCM_PATH"):