Skip to content

Commit 37b8bee

Browse files
sestinjsestinj
authored
Llama 3.1 405b model selection (#1813)
* add support for llama3 405b * 3.1 * name mapping * updates to model setup
1 parent be035fd commit 37b8bee

File tree

14 files changed

+407
-216
lines changed

14 files changed

+407
-216
lines changed

‎core/autocomplete/templates.ts

-5
Original file line numberDiff line numberDiff line change
@@ -155,11 +155,6 @@ const deepseekFimTemplate: AutocompleteTemplate = {
155155
},
156156
};
157157

158-
const deepseekFimTemplateWrongPipeChar: AutocompleteTemplate = {
159-
template: "<|fim▁begin|>{{{prefix}}}<|fim▁hole|>{{{suffix}}}<|fim▁end|>",
160-
completionOptions: { stop: ["<|fim▁begin|>", "<|fim▁hole|>", "<|fim▁end|>"] },
161-
};
162-
163158
const gptAutocompleteTemplate: AutocompleteTemplate = {
164159
template: `\`\`\`
165160
{{{prefix}}}[BLANK]{{{suffix}}}

‎core/llm/llms/FreeTrial.ts

+2-1
Original file line numberDiff line numberDiff line change
@@ -165,7 +165,8 @@ class FreeTrial extends BaseLLM {
165165
return [
166166
"codestral-latest",
167167
"claude-3-5-sonnet-20240620",
168-
"llama3-70b",
168+
"llama3.1-405b",
169+
"llama3.1-70b",
169170
"gpt-4o",
170171
"gpt-3.5-turbo",
171172
"claude-3-haiku-20240307",

‎core/llm/llms/Groq.ts

+3
Original file line numberDiff line numberDiff line change
@@ -14,6 +14,9 @@ class Groq extends OpenAI {
1414
gemma: "gemma-7b-it",
1515
"llama3-8b": "llama3-8b-8192",
1616
"llama3-70b": "llama3-70b-8192",
17+
"llama3.1-8b": "llama-3.1-8b-instant",
18+
"llama3.1-70b": "llama-3.1-70b-versatile",
19+
"llama3.1-405b": "llama-3.1-405b-reasoning",
1720
};
1821
protected _convertModelName(model: string): string {
1922
return Groq.modelConversion[model] ?? model;

‎core/llm/llms/Ollama.ts

+3
Original file line numberDiff line numberDiff line change
@@ -88,6 +88,9 @@ class Ollama extends BaseLLM {
8888
"codellama-70b": "codellama:70b",
8989
"llama3-8b": "llama3:8b",
9090
"llama3-70b": "llama3:70b",
91+
"llama3.1-8b": "llama3.1:8b",
92+
"llama3.1-70b": "llama3.1:70b",
93+
"llama3.1-405b": "llama3.1:405b",
9194
"phi-2": "phi:2.7b",
9295
"phind-codellama-34b": "phind-codellama:34b-v2",
9396
"wizardcoder-7b": "wizardcoder:7b-python",

‎core/llm/llms/Replicate.ts

+3
Original file line numberDiff line numberDiff line change
@@ -18,6 +18,9 @@ class Replicate extends BaseLLM {
1818
"llama2-13b": "meta/llama-2-13b-chat" as any,
1919
"llama3-8b": "meta/meta-llama-3-8b-instruct" as any,
2020
"llama3-70b": "meta/meta-llama-3-70b-instruct" as any,
21+
"llama3.1-8b": "meta/meta-llama-3.1-8b-instruct" as any,
22+
"llama3.1-70b": "meta/meta-llama-3.1-70b-instruct" as any,
23+
"llama3.1-405b": "meta/meta-llama-3.1-405b-instruct" as any,
2124
"zephyr-7b":
2225
"nateraw/zephyr-7b-beta:b79f33de5c6c4e34087d44eaea4a9d98ce5d3f3a09522f7328eea0685003a931",
2326
"mistral-7b":

‎core/llm/llms/Together.ts

+3
Original file line numberDiff line numberDiff line change
@@ -19,6 +19,9 @@ class Together extends OpenAI {
1919
"codellama-70b": "codellama/CodeLlama-70b-Instruct-hf",
2020
"llama3-8b": "meta-llama/Llama-3-8b-chat-hf",
2121
"llama3-70b": "meta-llama/Llama-3-70b-chat-hf",
22+
"llama3.1-8b": "meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
23+
"llama3.1-70b": "meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo",
24+
"llama3.1-405b": "meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo",
2225
"llama2-7b": "togethercomputer/llama-2-7b-chat",
2326
"llama2-13b": "togethercomputer/llama-2-13b-chat",
2427
"llama2-70b": "togethercomputer/llama-2-70b-chat",

‎docs/static/schemas/config.json

+23-1
Original file line numberDiff line numberDiff line change
@@ -514,6 +514,9 @@
514514
"anyOf": [
515515
{
516516
"enum": [
517+
"llama3.1-8b",
518+
"llama3.1-70b",
519+
"llama3.1-405b",
517520
"llama3-8b",
518521
"llama3-70b",
519522
"codellama-7b",
@@ -547,7 +550,8 @@
547550
"enum": [
548551
"gpt-4o",
549552
"codestral-latest",
550-
"llama3-70b",
553+
"llama3.1-70b",
554+
"llama3.1-405b",
551555
"gpt-3.5-turbo",
552556
"phind-codellama-34b",
553557
"gemini-pro",
@@ -599,6 +603,9 @@
599603
"codellama-70b",
600604
"llama3-8b",
601605
"llama3-70b",
606+
"llama3.1-8b",
607+
"llama3.1-70b",
608+
"llama3.1-405b",
602609
"phind-codellama-34b",
603610
"wizardcoder-7b",
604611
"wizardcoder-13b",
@@ -749,6 +756,9 @@
749756
"llama2-13b",
750757
"llama3-8b",
751758
"llama3-70b",
759+
"llama3.1-8b",
760+
"llama3.1-70b",
761+
"llama3.1-405b",
752762
"codellama-7b",
753763
"codellama-13b",
754764
"codellama-34b",
@@ -813,6 +823,9 @@
813823
"codellama-70b",
814824
"llama3-8b",
815825
"llama3-70b",
826+
"llama3.1-8b",
827+
"llama3.1-70b",
828+
"llama3.1-405b",
816829
"phind-codellama-34b",
817830
"wizardcoder-7b",
818831
"wizardcoder-13b",
@@ -861,6 +874,9 @@
861874
"codellama-70b",
862875
"llama3-8b",
863876
"llama3-70b",
877+
"llama3.1-8b",
878+
"llama3.1-70b",
879+
"llama3.1-405b",
864880
"phi-2",
865881
"phind-codellama-34b",
866882
"wizardcoder-7b",
@@ -948,6 +964,9 @@
948964
"gemma",
949965
"llama3-8b",
950966
"llama3-70b",
967+
"llama3.1-8b",
968+
"llama3.1-70b",
969+
"llama3.1-405b",
951970
"AUTODETECT"
952971
]
953972
}
@@ -1114,6 +1133,9 @@
11141133
"codellama-70b",
11151134
"llama3-8b",
11161135
"llama3-70b",
1136+
"llama3.1-8b",
1137+
"llama3.1-70b",
1138+
"llama3.1-405b",
11171139
"phind-codellama-34b",
11181140
"wizardcoder-7b",
11191141
"wizardcoder-13b",

‎extensions/intellij/src/main/resources/config_schema.json

+23-1
Original file line numberDiff line numberDiff line change
@@ -513,6 +513,9 @@
513513
"anyOf": [
514514
{
515515
"enum": [
516+
"llama3.1-8b",
517+
"llama3.1-70b",
518+
"llama3.1-405b",
516519
"llama3-8b",
517520
"llama3-70b",
518521
"codellama-7b",
@@ -546,7 +549,8 @@
546549
"enum": [
547550
"gpt-4o",
548551
"codestral-latest",
549-
"llama3-70b",
552+
"llama3.1-70b",
553+
"llama3.1-405b",
550554
"gpt-3.5-turbo",
551555
"phind-codellama-34b",
552556
"gemini-pro",
@@ -598,6 +602,9 @@
598602
"codellama-70b",
599603
"llama3-8b",
600604
"llama3-70b",
605+
"llama3.1-8b",
606+
"llama3.1-70b",
607+
"llama3.1-405b",
601608
"phind-codellama-34b",
602609
"wizardcoder-7b",
603610
"wizardcoder-13b",
@@ -748,6 +755,9 @@
748755
"llama2-13b",
749756
"llama3-8b",
750757
"llama3-70b",
758+
"llama3.1-8b",
759+
"llama3.1-70b",
760+
"llama3.1-405b",
751761
"codellama-7b",
752762
"codellama-13b",
753763
"codellama-34b",
@@ -812,6 +822,9 @@
812822
"codellama-70b",
813823
"llama3-8b",
814824
"llama3-70b",
825+
"llama3.1-8b",
826+
"llama3.1-70b",
827+
"llama3.1-405b",
815828
"phind-codellama-34b",
816829
"wizardcoder-7b",
817830
"wizardcoder-13b",
@@ -860,6 +873,9 @@
860873
"codellama-70b",
861874
"llama3-8b",
862875
"llama3-70b",
876+
"llama3.1-8b",
877+
"llama3.1-70b",
878+
"llama3.1-405b",
863879
"phi-2",
864880
"phind-codellama-34b",
865881
"wizardcoder-7b",
@@ -947,6 +963,9 @@
947963
"gemma",
948964
"llama3-8b",
949965
"llama3-70b",
966+
"llama3.1-8b",
967+
"llama3.1-70b",
968+
"llama3.1-405b",
950969
"AUTODETECT"
951970
]
952971
}
@@ -1113,6 +1132,9 @@
11131132
"codellama-70b",
11141133
"llama3-8b",
11151134
"llama3-70b",
1135+
"llama3.1-8b",
1136+
"llama3.1-70b",
1137+
"llama3.1-405b",
11161138
"phind-codellama-34b",
11171139
"wizardcoder-7b",
11181140
"wizardcoder-13b",

‎extensions/vscode/config_schema.json

+23-1
Original file line numberDiff line numberDiff line change
@@ -513,6 +513,9 @@
513513
"anyOf": [
514514
{
515515
"enum": [
516+
"llama3.1-8b",
517+
"llama3.1-70b",
518+
"llama3.1-405b",
516519
"llama3-8b",
517520
"llama3-70b",
518521
"codellama-7b",
@@ -546,7 +549,8 @@
546549
"enum": [
547550
"gpt-4o",
548551
"codestral-latest",
549-
"llama3-70b",
552+
"llama3.1-70b",
553+
"llama3.1-405b",
550554
"gpt-3.5-turbo",
551555
"phind-codellama-34b",
552556
"gemini-pro",
@@ -598,6 +602,9 @@
598602
"codellama-70b",
599603
"llama3-8b",
600604
"llama3-70b",
605+
"llama3.1-8b",
606+
"llama3.1-70b",
607+
"llama3.1-405b",
601608
"phind-codellama-34b",
602609
"wizardcoder-7b",
603610
"wizardcoder-13b",
@@ -748,6 +755,9 @@
748755
"llama2-13b",
749756
"llama3-8b",
750757
"llama3-70b",
758+
"llama3.1-8b",
759+
"llama3.1-70b",
760+
"llama3.1-405b",
751761
"codellama-7b",
752762
"codellama-13b",
753763
"codellama-34b",
@@ -812,6 +822,9 @@
812822
"codellama-70b",
813823
"llama3-8b",
814824
"llama3-70b",
825+
"llama3.1-8b",
826+
"llama3.1-70b",
827+
"llama3.1-405b",
815828
"phind-codellama-34b",
816829
"wizardcoder-7b",
817830
"wizardcoder-13b",
@@ -860,6 +873,9 @@
860873
"codellama-70b",
861874
"llama3-8b",
862875
"llama3-70b",
876+
"llama3.1-8b",
877+
"llama3.1-70b",
878+
"llama3.1-405b",
863879
"phi-2",
864880
"phind-codellama-34b",
865881
"wizardcoder-7b",
@@ -947,6 +963,9 @@
947963
"gemma",
948964
"llama3-8b",
949965
"llama3-70b",
966+
"llama3.1-8b",
967+
"llama3.1-70b",
968+
"llama3.1-405b",
950969
"AUTODETECT"
951970
]
952971
}
@@ -1113,6 +1132,9 @@
11131132
"codellama-70b",
11141133
"llama3-8b",
11151134
"llama3-70b",
1135+
"llama3.1-8b",
1136+
"llama3.1-70b",
1137+
"llama3.1-405b",
11161138
"phind-codellama-34b",
11171139
"wizardcoder-7b",
11181140
"wizardcoder-13b",

‎extensions/vscode/continue_rc_schema.json

+23-1
Original file line numberDiff line numberDiff line change
@@ -552,6 +552,9 @@
552552
"anyOf": [
553553
{
554554
"enum": [
555+
"llama3.1-8b",
556+
"llama3.1-70b",
557+
"llama3.1-405b",
555558
"llama3-8b",
556559
"llama3-70b",
557560
"codellama-7b",
@@ -589,7 +592,8 @@
589592
"enum": [
590593
"gpt-4o",
591594
"codestral-latest",
592-
"llama3-70b",
595+
"llama3.1-70b",
596+
"llama3.1-405b",
593597
"gpt-3.5-turbo",
594598
"phind-codellama-34b",
595599
"gemini-pro",
@@ -645,6 +649,9 @@
645649
"codellama-70b",
646650
"llama3-8b",
647651
"llama3-70b",
652+
"llama3.1-8b",
653+
"llama3.1-70b",
654+
"llama3.1-405b",
648655
"phind-codellama-34b",
649656
"wizardcoder-7b",
650657
"wizardcoder-13b",
@@ -818,6 +825,9 @@
818825
"llama2-13b",
819826
"llama3-8b",
820827
"llama3-70b",
828+
"llama3.1-8b",
829+
"llama3.1-70b",
830+
"llama3.1-405b",
821831
"codellama-7b",
822832
"codellama-13b",
823833
"codellama-34b",
@@ -888,6 +898,9 @@
888898
"codellama-70b",
889899
"llama3-8b",
890900
"llama3-70b",
901+
"llama3.1-8b",
902+
"llama3.1-70b",
903+
"llama3.1-405b",
891904
"phind-codellama-34b",
892905
"wizardcoder-7b",
893906
"wizardcoder-13b",
@@ -940,6 +953,9 @@
940953
"codellama-70b",
941954
"llama3-8b",
942955
"llama3-70b",
956+
"llama3.1-8b",
957+
"llama3.1-70b",
958+
"llama3.1-405b",
943959
"phi-2",
944960
"phind-codellama-34b",
945961
"wizardcoder-7b",
@@ -1042,6 +1058,9 @@
10421058
"gemma",
10431059
"llama3-8b",
10441060
"llama3-70b",
1061+
"llama3.1-8b",
1062+
"llama3.1-70b",
1063+
"llama3.1-405b",
10451064
"AUTODETECT"
10461065
]
10471066
}
@@ -1240,6 +1259,9 @@
12401259
"codellama-70b",
12411260
"llama3-8b",
12421261
"llama3-70b",
1262+
"llama3.1-8b",
1263+
"llama3.1-70b",
1264+
"llama3.1-405b",
12431265
"phind-codellama-34b",
12441266
"wizardcoder-7b",
12451267
"wizardcoder-13b",

‎extensions/vscode/package-lock.json

+2-2
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

0 commit comments

Comments
 (0)