[ChatQnA] Update OOB with wrapper manifests. (#823)

opea-project · Sep 19, 2024 · 933c3d3 · 933c3d3
1 parent 88829c9
commit 933c3d3
Show file tree

Hide file tree

Showing 6 changed files with 0 additions and 24 deletions.
diff --git a/ChatQnA/benchmark/oob/with_rerank/four_gaudi/oob_four_gaudi_with_rerank.yaml b/ChatQnA/benchmark/oob/with_rerank/four_gaudi/oob_four_gaudi_with_rerank.yaml
@@ -275,10 +275,6 @@ spec:
  - '2048'
  - --max-total-tokens
  - '4096'
- - --max-batch-total-tokens
- - '65536'
- - --max-batch-prefill-tokens
- - '4096'
  env:
  - name: OMPI_MCA_btl_vader_single_copy_mechanism
  value: none

diff --git a/ChatQnA/benchmark/oob/with_rerank/single_gaudi/oob_single_gaudi_with_rerank.yaml b/ChatQnA/benchmark/oob/with_rerank/single_gaudi/oob_single_gaudi_with_rerank.yaml
@@ -275,10 +275,6 @@ spec:
  - '2048'
  - --max-total-tokens
  - '4096'
- - --max-batch-total-tokens
- - '65536'
- - --max-batch-prefill-tokens
- - '4096'
  env:
  - name: OMPI_MCA_btl_vader_single_copy_mechanism
  value: none

diff --git a/ChatQnA/benchmark/oob/with_rerank/two_gaudi/oob_two_gaudi_with_rerank.yaml b/ChatQnA/benchmark/oob/with_rerank/two_gaudi/oob_two_gaudi_with_rerank.yaml
@@ -275,10 +275,6 @@ spec:
  - '2048'
  - --max-total-tokens
  - '4096'
- - --max-batch-total-tokens
- - '65536'
- - --max-batch-prefill-tokens
- - '4096'
  env:
  - name: OMPI_MCA_btl_vader_single_copy_mechanism
  value: none

diff --git a/ChatQnA/benchmark/oob/without_rerank/four_gaudi/oob_four_gaudi_without_rerank.yaml b/ChatQnA/benchmark/oob/without_rerank/four_gaudi/oob_four_gaudi_without_rerank.yaml
@@ -336,10 +336,6 @@ spec:
  - '2048'
  - --max-total-tokens
  - '4096'
- - --max-batch-total-tokens
- - '65536'
- - --max-batch-prefill-tokens
- - '4096'
  volumeMounts:
  - mountPath: /data
  name: model-volume

diff --git a/ChatQnA/benchmark/oob/without_rerank/single_gaudi/oob_single_gaudi_without_rerank.yaml b/ChatQnA/benchmark/oob/without_rerank/single_gaudi/oob_single_gaudi_without_rerank.yaml
@@ -336,10 +336,6 @@ spec:
  - '2048'
  - --max-total-tokens
  - '4096'
- - --max-batch-total-tokens
- - '65536'
- - --max-batch-prefill-tokens
- - '4096'
  volumeMounts:
  - mountPath: /data
  name: model-volume

diff --git a/ChatQnA/benchmark/oob/without_rerank/two_gaudi/oob_two_gaudi_without_rerank.yaml b/ChatQnA/benchmark/oob/without_rerank/two_gaudi/oob_two_gaudi_without_rerank.yaml
@@ -336,10 +336,6 @@ spec:
  - '2048'
  - --max-total-tokens
  - '4096'
- - --max-batch-total-tokens
- - '65536'
- - --max-batch-prefill-tokens
- - '4096'
  volumeMounts:
  - mountPath: /data
  name: model-volume