roleplays, go up to 7 threads due to benchmark data. 8 is very marginal, leave it for the system. 4 is great for MoE tokens/sec but 7 is still a bit faster in overall response time.
This commit is contained in:
1 parent
6411d4792a
commit
74789ba70c
1 file changed
+9
-9
@@ -8,7 +8,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_ctx": 256000,
|
"num_ctx": 256000,
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"system": "Please use the attached Knowledge Base files as your personality."
|
"system": "Please use the attached Knowledge Base files as your personality."
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -77,7 +77,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_ctx": 256000,
|
"num_ctx": 256000,
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"system": "Please use the attached Knowledge Base files as your personality."
|
"system": "Please use the attached Knowledge Base files as your personality."
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -146,7 +146,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_ctx": 256000,
|
"num_ctx": 256000,
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"system": "Please use the attached Knowledge Base files as your personality."
|
"system": "Please use the attached Knowledge Base files as your personality."
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -215,7 +215,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"system": "Please use the attached Knowledge Base files as your personality.",
|
"system": "Please use the attached Knowledge Base files as your personality.",
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"num_ctx": 32000
|
"num_ctx": 32000
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -279,7 +279,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"system": "Please use the attached Knowledge Base files as your personality.",
|
"system": "Please use the attached Knowledge Base files as your personality.",
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"num_ctx": 32000
|
"num_ctx": 32000
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -343,7 +343,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"system": "Please use the attached Knowledge Base files as your personality.",
|
"system": "Please use the attached Knowledge Base files as your personality.",
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"num_ctx": 64000
|
"num_ctx": 64000
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -407,7 +407,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_ctx": 256000,
|
"num_ctx": 256000,
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"system": "Please use the attached Knowledge Base files as your personality."
|
"system": "Please use the attached Knowledge Base files as your personality."
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -476,7 +476,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_ctx": 192000,
|
"num_ctx": 192000,
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"system": "Please use the attached Knowledge Base files as your personality."
|
"system": "Please use the attached Knowledge Base files as your personality."
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
@@ -545,7 +545,7 @@
|
|||||||
"params": {
|
"params": {
|
||||||
"system": "Please use the attached Knowledge Base files as your personality.",
|
"system": "Please use the attached Knowledge Base files as your personality.",
|
||||||
"custom_params": {},
|
"custom_params": {},
|
||||||
"num_thread": 6,
|
"num_thread": 7,
|
||||||
"num_ctx": 32000
|
"num_ctx": 32000
|
||||||
},
|
},
|
||||||
"meta": {
|
"meta": {
|
||||||
|
|||||||
Reference in new issue
Block a user