[Config Refactor] Derive deploy override fields from stage config (#3162)
Signed-off-by: xiaohajiayou <923390377@qq.com>
This commit is contained in:
15 files changed
+129
-59
No files matched your search
@@ -40,7 +40,7 @@ stages:
|
||||
# near-identity repetition penalty forces vLLM to track
|
||||
# output_token_ids for RAS (stop-token logit logsumexp).
|
||||
repetition_penalty: 1.0001
|
||||
disable_hybrid_kv_cache_manager: true
|
||||
enable_prefix_caching: false
|
||||
mm_processor_cache_gb: 0
|
||||
skip_mm_profiling: true
|
||||
|
||||
@@ -54,5 +54,5 @@ stages:
|
||||
from_stage_0: connector_of_shared_memory
|
||||
default_sampling_params:
|
||||
max_tokens: 2048
|
||||
disable_hybrid_kv_cache_manager: true
|
||||
enable_prefix_caching: false
|
||||
skip_mm_profiling: true
|
||||
@@ -42,6 +42,7 @@ stages:
|
||||
max_num_seqs: 1
|
||||
gpu_memory_utilization: 0.2
|
||||
enforce_eager: true
|
||||
enable_prefix_caching: false
|
||||
async_scheduling: false
|
||||
max_num_batched_tokens: 8192
|
||||
max_model_len: 8192
|
||||
|
||||
@@ -22,6 +22,7 @@ stages:
|
||||
max_num_seqs: 1
|
||||
gpu_memory_utilization: 0.8
|
||||
enforce_eager: true
|
||||
enable_prefix_caching: false
|
||||
mm_processor_cache_gb: 0
|
||||
devices: "0"
|
||||
default_sampling_params:
|
||||
@@ -49,6 +50,7 @@ stages:
|
||||
max_num_seqs: 1
|
||||
gpu_memory_utilization: 0.15
|
||||
enforce_eager: true
|
||||
enable_prefix_caching: false
|
||||
enable_flashinfer_autotune: false
|
||||
async_scheduling: false
|
||||
devices: "0"
|
||||
|
||||
@@ -23,6 +23,7 @@ connectors:
|
||||
stages:
|
||||
- stage_id: 0
|
||||
gpu_memory_utilization: 0.9
|
||||
enable_prefix_caching: false
|
||||
devices: "0"
|
||||
default_sampling_params:
|
||||
temperature: 0.4
|
||||
@@ -47,6 +48,7 @@ stages:
|
||||
- stage_id: 2
|
||||
gpu_memory_utilization: 0.1
|
||||
enforce_eager: true
|
||||
enable_prefix_caching: false
|
||||
async_scheduling: false
|
||||
max_num_batched_tokens: 51200
|
||||
devices: "1"
|
||||
|
||||
@@ -53,6 +53,7 @@ stages:
|
||||
max_num_seqs: 1
|
||||
gpu_memory_utilization: 0.3
|
||||
enforce_eager: true
|
||||
enable_prefix_caching: false
|
||||
async_scheduling: true
|
||||
# Must be divisible by num_code_groups and cover (left_context + chunk).
|
||||
# Prefill length is Q * num_frames (e.g. 16 * 2148 = 34368); keep
|
||||
|
||||
@@ -20,6 +20,7 @@ stages:
|
||||
max_num_batched_tokens: 4096
|
||||
max_model_len: 4096
|
||||
devices: "0"
|
||||
trust_remote_code: true
|
||||
default_sampling_params:
|
||||
temperature: 0.0
|
||||
top_p: 1.0
|
||||
|
||||
@@ -24,6 +24,7 @@ stages:
|
||||
max_num_seqs: 32
|
||||
gpu_memory_utilization: 0.8
|
||||
enforce_eager: false
|
||||
enable_prefix_caching: false
|
||||
async_scheduling: true
|
||||
max_model_len: 4096
|
||||
devices: "0"
|
||||
@@ -48,6 +49,7 @@ stages:
|
||||
max_num_seqs: 32
|
||||
gpu_memory_utilization: 0.1
|
||||
enforce_eager: true
|
||||
enable_prefix_caching: false
|
||||
async_scheduling: false
|
||||
max_num_batched_tokens: 65536
|
||||
max_model_len: 65536
|
||||
|
||||
Reference in new issue
Block a user