enable MTP on MTP models
This commit is contained in:
@@ -190,6 +190,8 @@ models:
|
|||||||
-hf unsloth/Qwen3.6-35B-A3B-MTP-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.6-35B-A3B-MTP-GGUF:Q4_K_M
|
||||||
${ctx_256k}
|
${ctx_256k}
|
||||||
${qwen35_think_args}
|
${qwen35_think_args}
|
||||||
|
--spec-type draft-mtp --spec-draft-n-max 1
|
||||||
|
--parallel 1
|
||||||
${common_args}
|
${common_args}
|
||||||
|
|
||||||
"unsloth/Qwen3.6-35B-A3B-MTP-GGUF-nothink:Q4_K_M":
|
"unsloth/Qwen3.6-35B-A3B-MTP-GGUF-nothink:Q4_K_M":
|
||||||
@@ -198,4 +200,6 @@ models:
|
|||||||
-hf unsloth/Qwen3.6-35B-A3B-MTP-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.6-35B-A3B-MTP-GGUF:Q4_K_M
|
||||||
${ctx_256k}
|
${ctx_256k}
|
||||||
${qwen35_nothink_args}
|
${qwen35_nothink_args}
|
||||||
|
--spec-type draft-mtp --spec-draft-n-max 1
|
||||||
|
--parallel 1
|
||||||
${common_args}
|
${common_args}
|
||||||
|
|||||||
Reference in New Issue
Block a user