|
|
@@ -61,8 +61,9 @@ args = [
|
|
|
#'{"method":"dflash","model":"incoai/GLM-5.3-Flash-DFlash2","num_speculative_tokens":7}'""",
|
|
|
"--speculative-config", """
|
|
|
'{"method":"mtp","num_speculative_tokens":5}'""",
|
|
|
+ # No --load-format instanttensor: the production-1.2 image does not ship
|
|
|
+ # the instanttensor package (the 2x lane also loads via plain safetensors).
|
|
|
"--language-model-only",
|
|
|
- "--load-format", "instanttensor",
|
|
|
# image requests need the mm template shipped/in-wallpaper on aiden's rigs:
|
|
|
# "--chat-template", "<path-to>/chat_template_mm.jinja",
|
|
|
]
|
