ehuaa
ehuaa
> My rule of thumb is if your losses are > 1.0 for early [1-3] layers, calibration data is off or tokenizer is not properly configured. Each module in each...
> My rule of thumb is if your losses are > 1.0 for early [1-3] layers, calibration data is off or tokenizer is not properly configured. Each module in each...
> What is your PPL before and after quantization? My PPL before quantization on wiki2 is 5.334， while after quantization the PPL is 5.415, my model is a finetuned version...
> I think it comes down to just adding `sliding_window` to the call for `_prepare_4d_causal_attention_mask_for_sdpa` yes. Would you like to open a PR? Sure，and i'll open a PR later in...
> It should if the mask is correctly passed yeah. New sdpa has the `sliding_window` argument anyway. Not sure it was correctly prepared before, important PR: #29407 @ArthurZucker Did you...
> @Chillee @kit1980 Have you solved this problem? I also found the tensor are different not only the first token, but all the logits of the tokens are different. @Chillee...
> @ehuaa The issue was raised because the script wants to install torch==1.7.1+cu110 as it depends; this is configured via TORCH_BLADE_CI_BUILD_TORCH_VERSION, see https://github.com/alibaba/BladeDISC/blob/main/pytorch_blade/scripts/build_pytorch_blade.sh#L32. > > BladeDISC already supports torch 2.0;...
ERROR: @local_config_cuda//:enable_cuda :: Error loading option @local_config_cuda//:enable_cuda: no such package '@llvm-raw//utils/bazel': java.io.IOException: Error downloading [https://storage.googleapis.com/mirror.tensorflow.org/github.com/llvm/llvm-project/archive/8c712296fb75ff73db08f92444b35c438c01a405.tar.gz, https://github.com/llvm/llvm-project/archive/8c712296fb75ff73db08f92444b35c438c01a405.tar.gz] to /home/banach/.cache/bazel/_bazel_banach/73d137a07d4a9c12dceaec8145974e25/external/llvm-raw/temp14641944661529926033/8c712296fb75ff73db08f92444b35c438c01a405.tar.gz: Premature EOF Traceback (most recent call last): File "/home/banach/BladeDISC/pytorch_blade/setup.py", line 151, in...
**/tests/torchscript:since_1_14.graph.test FAILED in 0.8s /home/banach/.cache/bazel/_bazel_banach/73d137a07d4a9c12dceaec8145974e25/execroot/org_torch_blade/bazel-out/k8-dbg/testlogs/tests/torchscript/since_1_14.graph.test/test.log Executed 37 out of 37 tests: 36 tests pass and 1 fails locally.** There were tests whose specified size is too big. Use the --test_verbose_timeout_warnings...