(tvm) branch nightly updated (f60b08c9a4 -> b654852b15)

2024-07-15 Thread github-bot
This is an automated email from the ASF dual-hosted git repository. github-bot pushed a change to branch nightly in repository https://gitbox.apache.org/repos/asf/tvm.git from f60b08c9a4 [QoL][IR] Provide default constructor for NameSupply/GlobalVarSupply (#17135) add eeebcfa0ad

Re: [PR] [Relay][Pytorch] Add support for `aten::scaled_dot_product_attention` [tvm]

2024-07-15 Thread via GitHub
yogurfrul commented on PR #16143: URL: https://github.com/apache/tvm/pull/16143#issuecomment-2229942936 when I run testcase in torch2.1 got error `IndexError: Dimension out of range (expected to be in range of [-3, 2], but got 3) ` -- This is an automated message from the Apache

[PR] [KVCache] PagedKVCache Quantization [tvm]

2024-07-15 Thread via GitHub
davidpissarra opened a new pull request, #17159: URL: https://github.com/apache/tvm/pull/17159 PR supporting https://github.com/mlc-ai/mlc-llm/pull/2663. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above

Re: [PR] [Relax] Implement R.ensure_aligned and update memory planning for R.view [tvm]

2024-07-15 Thread via GitHub
vinx13 commented on code in PR #17145: URL: https://github.com/apache/tvm/pull/17145#discussion_r1678444377 ## src/relax/op/memory/view.cc: ## @@ -334,13 +334,12 @@ Expr LegalizeView(const BlockBuilder& bb, const Call& call) { relative_byte_offset =

Re: [PR] [Relax] Integrate cuDNN attention [tvm]

2024-07-15 Thread via GitHub
vinx13 commented on code in PR #17157: URL: https://github.com/apache/tvm/pull/17157#discussion_r1678431685 ## python/tvm/contrib/cutlass/gen_tensor_op.py: ## @@ -745,8 +745,8 @@ def get_batch_on_arg(arg_name, arg_shape): attrs["qkv"] = func_args[0]

Re: [PR] [Relax] Integrate cuDNN attention [tvm]

2024-07-15 Thread via GitHub
vinx13 commented on PR #17157: URL: https://github.com/apache/tvm/pull/17157#issuecomment-2229492891 The new attention can be applied via cudnn BYOC. The decision of which BYOC backend (cudnn, cutlass) to use is left to the users. cudnn is likely to perform better on H100 as it has

Re: [PR] [CI] Remove lint step from `unity/pr-head` step [tvm]

2024-07-15 Thread via GitHub
Lunderberg commented on PR #17155: URL: https://github.com/apache/tvm/pull/17155#issuecomment-2229433972 @tvm-bot rerun -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To

[PR] [TIR][Analyzer] Simplify `x==x` expressions for all dtypes [tvm]

2024-07-15 Thread via GitHub
Lunderberg opened a new pull request, #17158: URL: https://github.com/apache/tvm/pull/17158 Prior to this commit, there was no rule to simplify `x == x` into `True`. In some cases, despite not having an explicit rewrite rule in `RewriteSimplifier`, the `RewriteSimplifier::CanProve`

Re: [PR] [Relax] Integrate cuDNN attention [tvm]

2024-07-15 Thread via GitHub
sunggg commented on code in PR #17157: URL: https://github.com/apache/tvm/pull/17157#discussion_r1678353644 ## python/tvm/contrib/cutlass/gen_tensor_op.py: ## @@ -745,8 +745,8 @@ def get_batch_on_arg(arg_name, arg_shape): attrs["qkv"] = func_args[0]

(tvm) branch main updated: [Bugfix] Allow import of TVM when current directory is read-only (#17142)

2024-07-15 Thread wuwei
This is an automated email from the ASF dual-hosted git repository. wuwei pushed a commit to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git The following commit(s) were added to refs/heads/main by this push: new b654852b15 [Bugfix] Allow import of TVM when current

Re: [PR] [Bugfix] Allow import of TVM when current directory is read-only [tvm]

2024-07-15 Thread via GitHub
vinx13 merged PR #17142: URL: https://github.com/apache/tvm/pull/17142 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail:

Re: [PR] [release][Dont Squash] Update version to 0.17.0 and 0.18.0.dev on main branch [tvm]

2024-07-15 Thread via GitHub
vinx13 merged PR #17156: URL: https://github.com/apache/tvm/pull/17156 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail:

(tvm) branch main updated (f60b08c9a4 -> 9a9386de08)

2024-07-15 Thread wuwei
This is an automated email from the ASF dual-hosted git repository. wuwei pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git from f60b08c9a4 [QoL][IR] Provide default constructor for NameSupply/GlobalVarSupply (#17135) add eeebcfa0ad [release]

Re: [PR] [Relax] Implement Rewriter class for pattern-rewrite [tvm]

2024-07-15 Thread via GitHub
Lunderberg commented on PR #17149: URL: https://github.com/apache/tvm/pull/17149#issuecomment-2228897543 > My main question for you is if this PR is aiming the full support for the existing pattern language. It is okay if there are some rough edges that we don't currently support, but I

Re: [PR] [Relax] Implement Rewriter class for pattern-rewrite [tvm]

2024-07-15 Thread via GitHub
Lunderberg commented on code in PR #17149: URL: https://github.com/apache/tvm/pull/17149#discussion_r1678069263 ## tests/python/relax/test_dataflow_rewriter.py: ## @@ -0,0 +1,1370 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license

[PR] [Relax] Integrate cuDNN attention [tvm]

2024-07-15 Thread via GitHub
vinx13 opened a new pull request, #17157: URL: https://github.com/apache/tvm/pull/17157 This integrates cuDNN attention kernels to BYOC. A dependency of cudnn_frontend is added. The cuDNN attention kernel supports fused qkv in BS3NH and SBN3H layouts. cc @sunggg @masahi @yongwww

Re: [PR] [Relax] Implement Rewriter class for pattern-rewrite [tvm]

2024-07-15 Thread via GitHub
Lunderberg commented on code in PR #17149: URL: https://github.com/apache/tvm/pull/17149#discussion_r1677910957 ## tests/python/relax/test_dataflow_rewriter.py: ## @@ -0,0 +1,1370 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license

Re: [PR] [Relax] Implement Rewriter class for pattern-rewrite [tvm]

2024-07-15 Thread via GitHub
Lunderberg commented on code in PR #17149: URL: https://github.com/apache/tvm/pull/17149#discussion_r1677904909 ## python/tvm/relax/dpl/rewrite.py: ## @@ -15,16 +15,154 @@ # specific language governing permissions and limitations # under the License. """APIs for

Re: [PR] [Relax] Implement Rewriter class for pattern-rewrite [tvm]

2024-07-15 Thread via GitHub
Lunderberg commented on code in PR #17149: URL: https://github.com/apache/tvm/pull/17149#discussion_r1677894422 ## python/tvm/relax/dpl/rewrite.py: ## @@ -15,16 +15,154 @@ # specific language governing permissions and limitations # under the License. """APIs for

Re: [PR] [Relax] Implement Rewriter class for pattern-rewrite [tvm]

2024-07-15 Thread via GitHub
Lunderberg commented on code in PR #17149: URL: https://github.com/apache/tvm/pull/17149#discussion_r1677870641 ## include/tvm/relax/block_builder.h: ## @@ -133,16 +133,47 @@ class BlockBuilderNode : public Object { * \brief Begin a new scope, with optional parameters that