Skip to content

Commit ddb1624

Browse files
CharlieFRuanCharlieFRuan
authored
[Model] Add SmolLM2, deprecate SmolLM1 (#623)
This PR adds the following SmolLM2 prebuilt models: - SmolLM2-1.7B-Instruct-q4f16_1-MLC - SmolLM2-1.7B-Instruct-q4f32_1-MLC - SmolLM2-360M-Instruct-q0f16-MLC - SmolLM2-360M-Instruct-q0f32-MLC - SmolLM2-360M-Instruct-q4f16_1-MLC - SmolLM2-360M-Instruct-q4f132_1-MLC - SmolLM2-135M-Instruct-q0f16-MLC - SmolLM2-135M-Instruct-q0f32-MLC Also removes SmolLM 1 models
1 parent 767e110 commit ddb1624

File tree

1 file changed

+41
-95
lines changed

1 file changed

+41
-95
lines changed

‎src/config.ts

Lines changed: 41 additions & 95 deletions
Original file line numberDiff line numberDiff line change
@@ -717,168 +717,114 @@ export const prebuiltAppConfig: AppConfig = {
717717
sliding_window_size: -1,
718718
},
719719
},
720-
// SmolLM
720+
// SmolLM2
721721
{
722-
model: "https://huggingface.co/mlc-ai/SmolLM-1.7B-Instruct-q0f16-MLC",
723-
model_id: "SmolLM-1.7B-Instruct-q0f16-MLC",
722+
model: "https://huggingface.co/mlc-ai/SmolLM2-1.7B-Instruct-q4f16_1-MLC",
723+
model_id: "SmolLM2-1.7B-Instruct-q4f16_1-MLC",
724724
model_lib:
725725
modelLibURLPrefix +
726726
modelVersion +
727-
"/SmolLM-1.7B-Instruct-q0f16-ctx2k_cs1k-webgpu.wasm",
728-
vram_required_MB: 3736.19,
727+
"/SmolLM2-1.7B-Instruct-q4f16_1-ctx4k_cs1k-webgpu.wasm",
728+
vram_required_MB: 1774.19,
729729
low_resource_required: true,
730730
required_features: ["shader-f16"],
731731
overrides: {
732-
context_window_size: 2048,
733-
},
734-
},
735-
{
736-
model: "https://huggingface.co/mlc-ai/SmolLM-1.7B-Instruct-q0f32-MLC",
737-
model_id: "SmolLM-1.7B-Instruct-q0f32-MLC",
738-
model_lib:
739-
modelLibURLPrefix +
740-
modelVersion +
741-
"/SmolLM-1.7B-Instruct-q0f32-ctx2k_cs1k-webgpu.wasm",
742-
vram_required_MB: 7432.38,
743-
low_resource_required: false,
744-
overrides: {
745-
context_window_size: 2048,
746-
},
747-
},
748-
{
749-
model: "https://huggingface.co/mlc-ai/SmolLM-1.7B-Instruct-q4f16_1-MLC",
750-
model_id: "SmolLM-1.7B-Instruct-q4f16_1-MLC",
751-
model_lib:
752-
modelLibURLPrefix +
753-
modelVersion +
754-
"/SmolLM-1.7B-Instruct-q4f16_1-ctx2k_cs1k-webgpu.wasm",
755-
vram_required_MB: 1390.19,
756-
low_resource_required: true,
757-
required_features: ["shader-f16"],
758-
overrides: {
759-
context_window_size: 2048,
732+
context_window_size: 4096,
760733
},
761734
},
762735
{
763-
model: "https://huggingface.co/mlc-ai/SmolLM-1.7B-Instruct-q4f32_1-MLC",
764-
model_id: "SmolLM-1.7B-Instruct-q4f32_1-MLC",
736+
model: "https://huggingface.co/mlc-ai/SmolLM2-1.7B-Instruct-q4f32_1-MLC",
737+
model_id: "SmolLM2-1.7B-Instruct-q4f32_1-MLC",
765738
model_lib:
766739
modelLibURLPrefix +
767740
modelVersion +
768-
"/SmolLM-1.7B-Instruct-q4f32_1-ctx2k_cs1k-webgpu.wasm",
769-
vram_required_MB: 1924.38,
741+
"/SmolLM2-1.7B-Instruct-q4f32_1-ctx4k_cs1k-webgpu.wasm",
742+
vram_required_MB: 2692.38,
770743
low_resource_required: true,
771744
overrides: {
772-
context_window_size: 2048,
745+
context_window_size: 4096,
773746
},
774747
},
775748

776749
{
777-
model: "https://huggingface.co/mlc-ai/SmolLM-360M-Instruct-q0f16-MLC",
778-
model_id: "SmolLM-360M-Instruct-q0f16-MLC",
779-
model_lib:
780-
modelLibURLPrefix +
781-
modelVersion +
782-
"/SmolLM-360M-Instruct-q0f16-ctx2k_cs1k-webgpu.wasm",
783-
vram_required_MB: 791.99,
784-
low_resource_required: true,
785-
required_features: ["shader-f16"],
786-
overrides: {
787-
context_window_size: 2048,
788-
},
789-
},
790-
{
791-
model: "https://huggingface.co/mlc-ai/SmolLM-360M-Instruct-q0f32-MLC",
792-
model_id: "SmolLM-360M-Instruct-q0f32-MLC",
750+
model: "https://huggingface.co/mlc-ai/SmolLM2-360M-Instruct-q0f16-MLC",
751+
model_id: "SmolLM2-360M-Instruct-q0f16-MLC",
793752
model_lib:
794753
modelLibURLPrefix +
795754
modelVersion +
796-
"/SmolLM-360M-Instruct-q0f32-ctx2k_cs1k-webgpu.wasm",
797-
vram_required_MB: 1583.99,
798-
low_resource_required: true,
799-
overrides: {
800-
context_window_size: 2048,
801-
},
802-
},
803-
{
804-
model: "https://huggingface.co/mlc-ai/SmolLM-360M-Instruct-q4f16_1-MLC",
805-
model_id: "SmolLM-360M-Instruct-q4f16_1-MLC",
806-
model_lib:
807-
modelLibURLPrefix +
808-
modelVersion +
809-
"/SmolLM-360M-Instruct-q4f16_1-ctx2k_cs1k-webgpu.wasm",
810-
vram_required_MB: 296.06,
755+
"/SmolLM2-360M-Instruct-q0f16-ctx4k_cs1k-webgpu.wasm",
756+
vram_required_MB: 871.99,
811757
low_resource_required: true,
812758
required_features: ["shader-f16"],
813759
overrides: {
814-
context_window_size: 2048,
760+
context_window_size: 4096,
815761
},
816762
},
817763
{
818-
model: "https://huggingface.co/mlc-ai/SmolLM-360M-Instruct-q4f32_1-MLC",
819-
model_id: "SmolLM-360M-Instruct-q4f32_1-MLC",
764+
model: "https://huggingface.co/mlc-ai/SmolLM2-360M-Instruct-q0f32-MLC",
765+
model_id: "SmolLM2-360M-Instruct-q0f32-MLC",
820766
model_lib:
821767
modelLibURLPrefix +
822768
modelVersion +
823-
"/SmolLM-360M-Instruct-q4f32_1-ctx2k_cs1k-webgpu.wasm",
824-
vram_required_MB: 419.61,
769+
"/SmolLM2-360M-Instruct-q0f32-ctx4k_cs1k-webgpu.wasm",
770+
vram_required_MB: 1743.99,
825771
low_resource_required: true,
826772
overrides: {
827-
context_window_size: 2048,
773+
context_window_size: 4096,
828774
},
829775
},
830776
{
831-
model: "https://huggingface.co/mlc-ai/SmolLM-135M-Instruct-q0f16-MLC",
832-
model_id: "SmolLM-135M-Instruct-q0f16-MLC",
777+
model: "https://huggingface.co/mlc-ai/SmolLM2-360M-Instruct-q4f16_1-MLC",
778+
model_id: "SmolLM2-360M-Instruct-q4f16_1-MLC",
833779
model_lib:
834780
modelLibURLPrefix +
835781
modelVersion +
836-
"/SmolLM-135M-Instruct-q0f16-ctx2k_cs1k-webgpu.wasm",
837-
vram_required_MB: 314.69,
782+
"/SmolLM2-360M-Instruct-q4f16_1-ctx4k_cs1k-webgpu.wasm",
783+
vram_required_MB: 376.06,
838784
low_resource_required: true,
839785
required_features: ["shader-f16"],
840786
overrides: {
841-
context_window_size: 2048,
787+
context_window_size: 4096,
842788
},
843789
},
844790
{
845-
model: "https://huggingface.co/mlc-ai/SmolLM-135M-Instruct-q0f32-MLC",
846-
model_id: "SmolLM-135M-Instruct-q0f32-MLC",
791+
model: "https://huggingface.co/mlc-ai/SmolLM2-360M-Instruct-q4f32_1-MLC",
792+
model_id: "SmolLM2-360M-Instruct-q4f32_1-MLC",
847793
model_lib:
848794
modelLibURLPrefix +
849795
modelVersion +
850-
"/SmolLM-135M-Instruct-q0f32-ctx2k_cs1k-webgpu.wasm",
851-
vram_required_MB: 629.38,
796+
"/SmolLM2-360M-Instruct-q4f32_1-ctx4k_cs1k-webgpu.wasm",
797+
vram_required_MB: 579.61,
852798
low_resource_required: true,
853799
overrides: {
854-
context_window_size: 2048,
800+
context_window_size: 4096,
855801
},
856802
},
857803
{
858-
model: "https://huggingface.co/mlc-ai/SmolLM-135M-Instruct-q4f16_1-MLC",
859-
model_id: "SmolLM-135M-Instruct-q4f16_1-MLC",
804+
model: "https://huggingface.co/mlc-ai/SmolLM2-135M-Instruct-q0f16-MLC",
805+
model_id: "SmolLM2-135M-Instruct-q0f16-MLC",
860806
model_lib:
861807
modelLibURLPrefix +
862808
modelVersion +
863-
"/SmolLM-135M-Instruct-q4f16_1-ctx2k_cs1k-webgpu.wasm",
864-
vram_required_MB: 130.33,
809+
"/SmolLM2-135M-Instruct-q0f16-ctx4k_cs1k-webgpu.wasm",
810+
vram_required_MB: 359.69,
865811
low_resource_required: true,
866812
required_features: ["shader-f16"],
867813
overrides: {
868-
context_window_size: 2048,
814+
context_window_size: 4096,
869815
},
870816
},
871817
{
872-
model: "https://huggingface.co/mlc-ai/SmolLM-135M-Instruct-q4f32_1-MLC",
873-
model_id: "SmolLM-135M-Instruct-q4f32_1-MLC",
818+
model: "https://huggingface.co/mlc-ai/SmolLM2-135M-Instruct-q0f32-MLC",
819+
model_id: "SmolLM2-135M-Instruct-q0f32-MLC",
874820
model_lib:
875821
modelLibURLPrefix +
876822
modelVersion +
877-
"/SmolLM-135M-Instruct-q4f32_1-ctx2k_cs1k-webgpu.wasm",
878-
vram_required_MB: 196.54,
823+
"/SmolLM2-135M-Instruct-q0f32-ctx4k_cs1k-webgpu.wasm",
824+
vram_required_MB: 719.38,
879825
low_resource_required: true,
880826
overrides: {
881-
context_window_size: 2048,
827+
context_window_size: 4096,
882828
},
883829
},
884830
// Gemma2

0 commit comments

Comments
 (0)