@infersec/conduit 1.111.0 → 1.112.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/cli.js +197 -1
- package/dist/cli.sea.cjs +197 -1
- package/package.json +1 -1
package/dist/cli.js
CHANGED
|
@@ -21567,6 +21567,90 @@ object$5({
|
|
|
21567
21567
|
// Source of truth for curated recommended models. Consumed by @infersec/definitions
|
|
21568
21568
|
// and the website model page generator (applications/website/scripts/generate-model-pages.ts).
|
|
21569
21569
|
const modelsData = [
|
|
21570
|
+
{
|
|
21571
|
+
author: "LiquidAI",
|
|
21572
|
+
baseModel: "LiquidAI/LFM2.5-2.6B",
|
|
21573
|
+
blurb: "Liquid's LFM2.5 2.6B hybrid at high fidelity - agentic post-training, 128K context, and 220 tok/s decode on an M5 Max. Fits a 4 GB card with room to spare.",
|
|
21574
|
+
contextLength: 131072,
|
|
21575
|
+
fileSizeBytes: 2874779648,
|
|
21576
|
+
format: "gguf",
|
|
21577
|
+
license: "lfm1.0",
|
|
21578
|
+
parameterCount: "2.6B",
|
|
21579
|
+
quantization: "Q8_0",
|
|
21580
|
+
slug: "LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21581
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21582
|
+
vramTierGB: 4
|
|
21583
|
+
},
|
|
21584
|
+
{
|
|
21585
|
+
author: "LiquidAI",
|
|
21586
|
+
baseModel: "LiquidAI/LFM2.5-2.6B",
|
|
21587
|
+
blurb: "Liquid's LFM2.5 2.6B hybrid at 4-bit - competitive with models 4x its size on tool use and multi-step agentic tasks. A 1.7 GB download that sips memory.",
|
|
21588
|
+
contextLength: 131072,
|
|
21589
|
+
fileSizeBytes: 1674455040,
|
|
21590
|
+
format: "gguf",
|
|
21591
|
+
license: "lfm1.0",
|
|
21592
|
+
parameterCount: "2.6B",
|
|
21593
|
+
quantization: "Q4_K_M",
|
|
21594
|
+
slug: "LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21595
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21596
|
+
vramTierGB: 4
|
|
21597
|
+
},
|
|
21598
|
+
{
|
|
21599
|
+
author: "LiquidAI",
|
|
21600
|
+
baseModel: "LiquidAI/LFM2.5-2.6B",
|
|
21601
|
+
blurb: "The native BF16 checkpoint of Liquid's LFM2.5 2.6B for vLLM and SGLang - hybrid conv + GQA architecture, 128K context window, and agentic reinforcement-learning post-training.",
|
|
21602
|
+
contextLength: 131072,
|
|
21603
|
+
fileSizeBytes: 5394427456,
|
|
21604
|
+
format: "safetensors",
|
|
21605
|
+
license: "lfm1.0",
|
|
21606
|
+
parameterCount: "2.6B",
|
|
21607
|
+
quantization: "BF16",
|
|
21608
|
+
slug: "LiquidAI/LFM2.5-2.6B",
|
|
21609
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-2.6B",
|
|
21610
|
+
vramTierGB: 8
|
|
21611
|
+
},
|
|
21612
|
+
{
|
|
21613
|
+
author: "LiquidAI",
|
|
21614
|
+
baseModel: "LiquidAI/LFM2.5-230M",
|
|
21615
|
+
blurb: "Liquid's most compact LFM2.5 at high fidelity - a 230M edge model distilled from the 350M, tuned for tool use and data extraction. 213 tok/s decode on a Galaxy S25 Ultra.",
|
|
21616
|
+
contextLength: 32768,
|
|
21617
|
+
fileSizeBytes: 246598496,
|
|
21618
|
+
format: "gguf",
|
|
21619
|
+
license: "lfm1.0",
|
|
21620
|
+
parameterCount: "230M",
|
|
21621
|
+
quantization: "Q8_0",
|
|
21622
|
+
slug: "LiquidAI/LFM2.5-230M-GGUF",
|
|
21623
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-230M-GGUF",
|
|
21624
|
+
vramTierGB: 2
|
|
21625
|
+
},
|
|
21626
|
+
{
|
|
21627
|
+
author: "LiquidAI",
|
|
21628
|
+
baseModel: "LiquidAI/LFM2.5-230M",
|
|
21629
|
+
blurb: "Liquid's most compact LFM2.5 at 4-bit - a 153 MB download that brings real tool-use capability to the tightest memory budgets, from Raspberry Pi to phones.",
|
|
21630
|
+
contextLength: 32768,
|
|
21631
|
+
fileSizeBytes: 153406304,
|
|
21632
|
+
format: "gguf",
|
|
21633
|
+
license: "lfm1.0",
|
|
21634
|
+
parameterCount: "230M",
|
|
21635
|
+
quantization: "Q4_K_M",
|
|
21636
|
+
slug: "LiquidAI/LFM2.5-230M-GGUF",
|
|
21637
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-230M-GGUF",
|
|
21638
|
+
vramTierGB: 2
|
|
21639
|
+
},
|
|
21640
|
+
{
|
|
21641
|
+
author: "LiquidAI",
|
|
21642
|
+
baseModel: "LiquidAI/LFM2.5-230M",
|
|
21643
|
+
blurb: "The native BF16 checkpoint of Liquid's LFM2.5 230M for vLLM - a 230M edge model with a 32K context window, suited to data extraction and lightweight on-device agent pipelines.",
|
|
21644
|
+
contextLength: 32768,
|
|
21645
|
+
fileSizeBytes: 459401112,
|
|
21646
|
+
format: "safetensors",
|
|
21647
|
+
license: "lfm1.0",
|
|
21648
|
+
parameterCount: "230M",
|
|
21649
|
+
quantization: "BF16",
|
|
21650
|
+
slug: "LiquidAI/LFM2.5-230M",
|
|
21651
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-230M",
|
|
21652
|
+
vramTierGB: 4
|
|
21653
|
+
},
|
|
21570
21654
|
{
|
|
21571
21655
|
author: "Qwen",
|
|
21572
21656
|
baseModel: "Qwen/Qwen3-8B",
|
|
@@ -21623,6 +21707,48 @@ const modelsData = [
|
|
|
21623
21707
|
sourceURL: "https://huggingface.co/bartowski/deepreinforce-ai_Ornith-1.0-9B-GGUF",
|
|
21624
21708
|
vramTierGB: 16
|
|
21625
21709
|
},
|
|
21710
|
+
{
|
|
21711
|
+
author: "unsloth",
|
|
21712
|
+
baseModel: "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
|
|
21713
|
+
blurb: "NVIDIA's Nemotron 3.5 Lightning MoE squeezed into 2-bit - 30B total with only 3B active in a Mamba-2 + MoE + attention hybrid. Configurable thinking and a 256K context on a single 24 GB card.",
|
|
21714
|
+
contextLength: 262144,
|
|
21715
|
+
fileSizeBytes: 19432070208,
|
|
21716
|
+
format: "gguf",
|
|
21717
|
+
license: "openmdw-1.1",
|
|
21718
|
+
parameterCount: "30B",
|
|
21719
|
+
quantization: "UD-IQ2_M",
|
|
21720
|
+
slug: "unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21721
|
+
sourceURL: "https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21722
|
+
vramTierGB: 24
|
|
21723
|
+
},
|
|
21724
|
+
{
|
|
21725
|
+
author: "unsloth",
|
|
21726
|
+
baseModel: "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
|
|
21727
|
+
blurb: "NVIDIA's Nemotron 3.5 Lightning at Unsloth's dynamic 4-bit - a Mamba-2 hybrid MoE with 3B active parameters, toggleable reasoning, and strong instruction following (IFBench 71.9). Needs a 32 GB GPU.",
|
|
21728
|
+
contextLength: 262144,
|
|
21729
|
+
fileSizeBytes: 25266255936,
|
|
21730
|
+
format: "gguf",
|
|
21731
|
+
license: "openmdw-1.1",
|
|
21732
|
+
parameterCount: "30B",
|
|
21733
|
+
quantization: "UD-Q4_K_M",
|
|
21734
|
+
slug: "unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21735
|
+
sourceURL: "https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21736
|
+
vramTierGB: 32
|
|
21737
|
+
},
|
|
21738
|
+
{
|
|
21739
|
+
author: "unsloth",
|
|
21740
|
+
baseModel: "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
|
|
21741
|
+
blurb: "Nemotron 3.5 Lightning at 8-bit fidelity - near-reference quality from the 30B-A3B Mamba-2 hybrid, with configurable thinking and DSpark speculative decoding support. Needs a 48 GB card.",
|
|
21742
|
+
contextLength: 262144,
|
|
21743
|
+
fileSizeBytes: 35004643392,
|
|
21744
|
+
format: "gguf",
|
|
21745
|
+
license: "openmdw-1.1",
|
|
21746
|
+
parameterCount: "30B",
|
|
21747
|
+
quantization: "Q8_0",
|
|
21748
|
+
slug: "unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21749
|
+
sourceURL: "https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21750
|
+
vramTierGB: 48
|
|
21751
|
+
},
|
|
21626
21752
|
{
|
|
21627
21753
|
author: "unsloth",
|
|
21628
21754
|
baseModel: "openai/gpt-oss-20b",
|
|
@@ -21681,6 +21807,76 @@ const modelsData = [
|
|
|
21681
21807
|
sourceURL: "https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF",
|
|
21682
21808
|
vramTierGB: 128
|
|
21683
21809
|
},
|
|
21810
|
+
{
|
|
21811
|
+
author: "Qwen",
|
|
21812
|
+
baseModel: "Qwen/Qwen3.8-Flash-Next",
|
|
21813
|
+
blurb: "The official BF16 release of Qwen's experimental Qwen4-preview architecture - 180B total with only 6B activated, plus 51B n-gram embeddings, a vision encoder, and a 262K native context. For vLLM/SGLang on multi-GPU server hardware.",
|
|
21814
|
+
contextLength: 262144,
|
|
21815
|
+
fileSizeBytes: 360000192888,
|
|
21816
|
+
format: "safetensors",
|
|
21817
|
+
license: "qwen-community-1.0",
|
|
21818
|
+
parameterCount: "180B",
|
|
21819
|
+
quantization: "BF16",
|
|
21820
|
+
slug: "Qwen/Qwen3.8-Flash-Next",
|
|
21821
|
+
sourceURL: "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
|
|
21822
|
+
vramTierGB: 640
|
|
21823
|
+
},
|
|
21824
|
+
{
|
|
21825
|
+
author: "RadixArk",
|
|
21826
|
+
baseModel: "Qwen/Qwen3.8-Flash-Next",
|
|
21827
|
+
blurb: "Qwen3.8-Flash-Next with its routed experts quantized to NVIDIA's NVFP4 W4A4 via Model Optimizer - 135 GB instead of 360 GB BF16, keeping GSM8K/AIME in-band with the reference. Serve with SGLang on Blackwell hardware.",
|
|
21828
|
+
contextLength: 262144,
|
|
21829
|
+
fileSizeBytes: 135195303851,
|
|
21830
|
+
format: "safetensors",
|
|
21831
|
+
license: "qwen-community-1.0",
|
|
21832
|
+
parameterCount: "180B",
|
|
21833
|
+
quantization: "NVFP4",
|
|
21834
|
+
slug: "RadixArk/Qwen3.8-Flash-Next-NVFP4",
|
|
21835
|
+
sourceURL: "https://huggingface.co/RadixArk/Qwen3.8-Flash-Next-NVFP4",
|
|
21836
|
+
vramTierGB: 256
|
|
21837
|
+
},
|
|
21838
|
+
{
|
|
21839
|
+
author: "deepseek-ai",
|
|
21840
|
+
baseModel: "deepseek-ai/DeepSeek-V4-Flash",
|
|
21841
|
+
blurb: "DeepSeek's V4-Flash - a 284B MoE with only 13B activated and a native 1M context window, shipped in mixed FP4/FP8 precision. Hybrid CSA/HCA attention keeps long-context inference cheap, with three reasoning-effort modes including Think Max.",
|
|
21842
|
+
contextLength: 1048576,
|
|
21843
|
+
fileSizeBytes: 159617149040,
|
|
21844
|
+
format: "safetensors",
|
|
21845
|
+
license: "mit",
|
|
21846
|
+
parameterCount: "284B",
|
|
21847
|
+
quantization: "FP4-FP8",
|
|
21848
|
+
slug: "deepseek-ai/DeepSeek-V4-Flash",
|
|
21849
|
+
sourceURL: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash",
|
|
21850
|
+
vramTierGB: 256
|
|
21851
|
+
},
|
|
21852
|
+
{
|
|
21853
|
+
author: "unsloth",
|
|
21854
|
+
baseModel: "deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
21855
|
+
blurb: "Unsloth's 2-bit cut of DeepSeek's experimental vision variant of V4-Flash - the same 284B MoE with 13B activated, plus image understanding via a separate mmproj projector. Fits 128 GB of unified memory or 2x 96 GB.",
|
|
21856
|
+
contextLength: 1048576,
|
|
21857
|
+
fileSizeBytes: 90692965818,
|
|
21858
|
+
format: "gguf",
|
|
21859
|
+
license: "mit",
|
|
21860
|
+
parameterCount: "284B",
|
|
21861
|
+
quantization: "UD-IQ2_XXS",
|
|
21862
|
+
slug: "unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21863
|
+
sourceURL: "https://huggingface.co/unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21864
|
+
vramTierGB: 128
|
|
21865
|
+
},
|
|
21866
|
+
{
|
|
21867
|
+
author: "unsloth",
|
|
21868
|
+
baseModel: "deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
21869
|
+
blurb: "The 4-bit cut of DeepSeek's experimental vision variant of V4-Flash - 284B MoE with 13B activated at higher fidelity, with multimodal agent scores well clear of the text-only release. Image input currently needs a preview llama.cpp build; needs 256 GB.",
|
|
21870
|
+
contextLength: 1048576,
|
|
21871
|
+
fileSizeBytes: 136662447706,
|
|
21872
|
+
format: "gguf",
|
|
21873
|
+
license: "mit",
|
|
21874
|
+
parameterCount: "284B",
|
|
21875
|
+
quantization: "UD-IQ4_XS",
|
|
21876
|
+
slug: "unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21877
|
+
sourceURL: "https://huggingface.co/unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21878
|
+
vramTierGB: 256
|
|
21879
|
+
},
|
|
21684
21880
|
{
|
|
21685
21881
|
author: "unsloth",
|
|
21686
21882
|
baseModel: "Qwen/Qwen3.6-27B",
|
|
@@ -342701,7 +342897,7 @@ async function runModelFit(options) {
|
|
|
342701
342897
|
? Math.max(...detection.gpus
|
|
342702
342898
|
.map(gpu => gpu.memoryTotalBytes ?? 0)
|
|
342703
342899
|
.filter(bytes => bytes > 0))
|
|
342704
|
-
: detection.memory.totalBytes)}. Smallest tier starts at
|
|
342900
|
+
: detection.memory.totalBytes)}. Smallest tier starts at 2 GB.`);
|
|
342705
342901
|
process.exitCode = 1;
|
|
342706
342902
|
return;
|
|
342707
342903
|
}
|
package/dist/cli.sea.cjs
CHANGED
|
@@ -21581,6 +21581,90 @@ object$5({
|
|
|
21581
21581
|
// Source of truth for curated recommended models. Consumed by @infersec/definitions
|
|
21582
21582
|
// and the website model page generator (applications/website/scripts/generate-model-pages.ts).
|
|
21583
21583
|
const modelsData = [
|
|
21584
|
+
{
|
|
21585
|
+
author: "LiquidAI",
|
|
21586
|
+
baseModel: "LiquidAI/LFM2.5-2.6B",
|
|
21587
|
+
blurb: "Liquid's LFM2.5 2.6B hybrid at high fidelity - agentic post-training, 128K context, and 220 tok/s decode on an M5 Max. Fits a 4 GB card with room to spare.",
|
|
21588
|
+
contextLength: 131072,
|
|
21589
|
+
fileSizeBytes: 2874779648,
|
|
21590
|
+
format: "gguf",
|
|
21591
|
+
license: "lfm1.0",
|
|
21592
|
+
parameterCount: "2.6B",
|
|
21593
|
+
quantization: "Q8_0",
|
|
21594
|
+
slug: "LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21595
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21596
|
+
vramTierGB: 4
|
|
21597
|
+
},
|
|
21598
|
+
{
|
|
21599
|
+
author: "LiquidAI",
|
|
21600
|
+
baseModel: "LiquidAI/LFM2.5-2.6B",
|
|
21601
|
+
blurb: "Liquid's LFM2.5 2.6B hybrid at 4-bit - competitive with models 4x its size on tool use and multi-step agentic tasks. A 1.7 GB download that sips memory.",
|
|
21602
|
+
contextLength: 131072,
|
|
21603
|
+
fileSizeBytes: 1674455040,
|
|
21604
|
+
format: "gguf",
|
|
21605
|
+
license: "lfm1.0",
|
|
21606
|
+
parameterCount: "2.6B",
|
|
21607
|
+
quantization: "Q4_K_M",
|
|
21608
|
+
slug: "LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21609
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-2.6B-GGUF",
|
|
21610
|
+
vramTierGB: 4
|
|
21611
|
+
},
|
|
21612
|
+
{
|
|
21613
|
+
author: "LiquidAI",
|
|
21614
|
+
baseModel: "LiquidAI/LFM2.5-2.6B",
|
|
21615
|
+
blurb: "The native BF16 checkpoint of Liquid's LFM2.5 2.6B for vLLM and SGLang - hybrid conv + GQA architecture, 128K context window, and agentic reinforcement-learning post-training.",
|
|
21616
|
+
contextLength: 131072,
|
|
21617
|
+
fileSizeBytes: 5394427456,
|
|
21618
|
+
format: "safetensors",
|
|
21619
|
+
license: "lfm1.0",
|
|
21620
|
+
parameterCount: "2.6B",
|
|
21621
|
+
quantization: "BF16",
|
|
21622
|
+
slug: "LiquidAI/LFM2.5-2.6B",
|
|
21623
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-2.6B",
|
|
21624
|
+
vramTierGB: 8
|
|
21625
|
+
},
|
|
21626
|
+
{
|
|
21627
|
+
author: "LiquidAI",
|
|
21628
|
+
baseModel: "LiquidAI/LFM2.5-230M",
|
|
21629
|
+
blurb: "Liquid's most compact LFM2.5 at high fidelity - a 230M edge model distilled from the 350M, tuned for tool use and data extraction. 213 tok/s decode on a Galaxy S25 Ultra.",
|
|
21630
|
+
contextLength: 32768,
|
|
21631
|
+
fileSizeBytes: 246598496,
|
|
21632
|
+
format: "gguf",
|
|
21633
|
+
license: "lfm1.0",
|
|
21634
|
+
parameterCount: "230M",
|
|
21635
|
+
quantization: "Q8_0",
|
|
21636
|
+
slug: "LiquidAI/LFM2.5-230M-GGUF",
|
|
21637
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-230M-GGUF",
|
|
21638
|
+
vramTierGB: 2
|
|
21639
|
+
},
|
|
21640
|
+
{
|
|
21641
|
+
author: "LiquidAI",
|
|
21642
|
+
baseModel: "LiquidAI/LFM2.5-230M",
|
|
21643
|
+
blurb: "Liquid's most compact LFM2.5 at 4-bit - a 153 MB download that brings real tool-use capability to the tightest memory budgets, from Raspberry Pi to phones.",
|
|
21644
|
+
contextLength: 32768,
|
|
21645
|
+
fileSizeBytes: 153406304,
|
|
21646
|
+
format: "gguf",
|
|
21647
|
+
license: "lfm1.0",
|
|
21648
|
+
parameterCount: "230M",
|
|
21649
|
+
quantization: "Q4_K_M",
|
|
21650
|
+
slug: "LiquidAI/LFM2.5-230M-GGUF",
|
|
21651
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-230M-GGUF",
|
|
21652
|
+
vramTierGB: 2
|
|
21653
|
+
},
|
|
21654
|
+
{
|
|
21655
|
+
author: "LiquidAI",
|
|
21656
|
+
baseModel: "LiquidAI/LFM2.5-230M",
|
|
21657
|
+
blurb: "The native BF16 checkpoint of Liquid's LFM2.5 230M for vLLM - a 230M edge model with a 32K context window, suited to data extraction and lightweight on-device agent pipelines.",
|
|
21658
|
+
contextLength: 32768,
|
|
21659
|
+
fileSizeBytes: 459401112,
|
|
21660
|
+
format: "safetensors",
|
|
21661
|
+
license: "lfm1.0",
|
|
21662
|
+
parameterCount: "230M",
|
|
21663
|
+
quantization: "BF16",
|
|
21664
|
+
slug: "LiquidAI/LFM2.5-230M",
|
|
21665
|
+
sourceURL: "https://huggingface.co/LiquidAI/LFM2.5-230M",
|
|
21666
|
+
vramTierGB: 4
|
|
21667
|
+
},
|
|
21584
21668
|
{
|
|
21585
21669
|
author: "Qwen",
|
|
21586
21670
|
baseModel: "Qwen/Qwen3-8B",
|
|
@@ -21637,6 +21721,48 @@ const modelsData = [
|
|
|
21637
21721
|
sourceURL: "https://huggingface.co/bartowski/deepreinforce-ai_Ornith-1.0-9B-GGUF",
|
|
21638
21722
|
vramTierGB: 16
|
|
21639
21723
|
},
|
|
21724
|
+
{
|
|
21725
|
+
author: "unsloth",
|
|
21726
|
+
baseModel: "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
|
|
21727
|
+
blurb: "NVIDIA's Nemotron 3.5 Lightning MoE squeezed into 2-bit - 30B total with only 3B active in a Mamba-2 + MoE + attention hybrid. Configurable thinking and a 256K context on a single 24 GB card.",
|
|
21728
|
+
contextLength: 262144,
|
|
21729
|
+
fileSizeBytes: 19432070208,
|
|
21730
|
+
format: "gguf",
|
|
21731
|
+
license: "openmdw-1.1",
|
|
21732
|
+
parameterCount: "30B",
|
|
21733
|
+
quantization: "UD-IQ2_M",
|
|
21734
|
+
slug: "unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21735
|
+
sourceURL: "https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21736
|
+
vramTierGB: 24
|
|
21737
|
+
},
|
|
21738
|
+
{
|
|
21739
|
+
author: "unsloth",
|
|
21740
|
+
baseModel: "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
|
|
21741
|
+
blurb: "NVIDIA's Nemotron 3.5 Lightning at Unsloth's dynamic 4-bit - a Mamba-2 hybrid MoE with 3B active parameters, toggleable reasoning, and strong instruction following (IFBench 71.9). Needs a 32 GB GPU.",
|
|
21742
|
+
contextLength: 262144,
|
|
21743
|
+
fileSizeBytes: 25266255936,
|
|
21744
|
+
format: "gguf",
|
|
21745
|
+
license: "openmdw-1.1",
|
|
21746
|
+
parameterCount: "30B",
|
|
21747
|
+
quantization: "UD-Q4_K_M",
|
|
21748
|
+
slug: "unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21749
|
+
sourceURL: "https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21750
|
+
vramTierGB: 32
|
|
21751
|
+
},
|
|
21752
|
+
{
|
|
21753
|
+
author: "unsloth",
|
|
21754
|
+
baseModel: "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
|
|
21755
|
+
blurb: "Nemotron 3.5 Lightning at 8-bit fidelity - near-reference quality from the 30B-A3B Mamba-2 hybrid, with configurable thinking and DSpark speculative decoding support. Needs a 48 GB card.",
|
|
21756
|
+
contextLength: 262144,
|
|
21757
|
+
fileSizeBytes: 35004643392,
|
|
21758
|
+
format: "gguf",
|
|
21759
|
+
license: "openmdw-1.1",
|
|
21760
|
+
parameterCount: "30B",
|
|
21761
|
+
quantization: "Q8_0",
|
|
21762
|
+
slug: "unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21763
|
+
sourceURL: "https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
|
|
21764
|
+
vramTierGB: 48
|
|
21765
|
+
},
|
|
21640
21766
|
{
|
|
21641
21767
|
author: "unsloth",
|
|
21642
21768
|
baseModel: "openai/gpt-oss-20b",
|
|
@@ -21695,6 +21821,76 @@ const modelsData = [
|
|
|
21695
21821
|
sourceURL: "https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF",
|
|
21696
21822
|
vramTierGB: 128
|
|
21697
21823
|
},
|
|
21824
|
+
{
|
|
21825
|
+
author: "Qwen",
|
|
21826
|
+
baseModel: "Qwen/Qwen3.8-Flash-Next",
|
|
21827
|
+
blurb: "The official BF16 release of Qwen's experimental Qwen4-preview architecture - 180B total with only 6B activated, plus 51B n-gram embeddings, a vision encoder, and a 262K native context. For vLLM/SGLang on multi-GPU server hardware.",
|
|
21828
|
+
contextLength: 262144,
|
|
21829
|
+
fileSizeBytes: 360000192888,
|
|
21830
|
+
format: "safetensors",
|
|
21831
|
+
license: "qwen-community-1.0",
|
|
21832
|
+
parameterCount: "180B",
|
|
21833
|
+
quantization: "BF16",
|
|
21834
|
+
slug: "Qwen/Qwen3.8-Flash-Next",
|
|
21835
|
+
sourceURL: "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
|
|
21836
|
+
vramTierGB: 640
|
|
21837
|
+
},
|
|
21838
|
+
{
|
|
21839
|
+
author: "RadixArk",
|
|
21840
|
+
baseModel: "Qwen/Qwen3.8-Flash-Next",
|
|
21841
|
+
blurb: "Qwen3.8-Flash-Next with its routed experts quantized to NVIDIA's NVFP4 W4A4 via Model Optimizer - 135 GB instead of 360 GB BF16, keeping GSM8K/AIME in-band with the reference. Serve with SGLang on Blackwell hardware.",
|
|
21842
|
+
contextLength: 262144,
|
|
21843
|
+
fileSizeBytes: 135195303851,
|
|
21844
|
+
format: "safetensors",
|
|
21845
|
+
license: "qwen-community-1.0",
|
|
21846
|
+
parameterCount: "180B",
|
|
21847
|
+
quantization: "NVFP4",
|
|
21848
|
+
slug: "RadixArk/Qwen3.8-Flash-Next-NVFP4",
|
|
21849
|
+
sourceURL: "https://huggingface.co/RadixArk/Qwen3.8-Flash-Next-NVFP4",
|
|
21850
|
+
vramTierGB: 256
|
|
21851
|
+
},
|
|
21852
|
+
{
|
|
21853
|
+
author: "deepseek-ai",
|
|
21854
|
+
baseModel: "deepseek-ai/DeepSeek-V4-Flash",
|
|
21855
|
+
blurb: "DeepSeek's V4-Flash - a 284B MoE with only 13B activated and a native 1M context window, shipped in mixed FP4/FP8 precision. Hybrid CSA/HCA attention keeps long-context inference cheap, with three reasoning-effort modes including Think Max.",
|
|
21856
|
+
contextLength: 1048576,
|
|
21857
|
+
fileSizeBytes: 159617149040,
|
|
21858
|
+
format: "safetensors",
|
|
21859
|
+
license: "mit",
|
|
21860
|
+
parameterCount: "284B",
|
|
21861
|
+
quantization: "FP4-FP8",
|
|
21862
|
+
slug: "deepseek-ai/DeepSeek-V4-Flash",
|
|
21863
|
+
sourceURL: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash",
|
|
21864
|
+
vramTierGB: 256
|
|
21865
|
+
},
|
|
21866
|
+
{
|
|
21867
|
+
author: "unsloth",
|
|
21868
|
+
baseModel: "deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
21869
|
+
blurb: "Unsloth's 2-bit cut of DeepSeek's experimental vision variant of V4-Flash - the same 284B MoE with 13B activated, plus image understanding via a separate mmproj projector. Fits 128 GB of unified memory or 2x 96 GB.",
|
|
21870
|
+
contextLength: 1048576,
|
|
21871
|
+
fileSizeBytes: 90692965818,
|
|
21872
|
+
format: "gguf",
|
|
21873
|
+
license: "mit",
|
|
21874
|
+
parameterCount: "284B",
|
|
21875
|
+
quantization: "UD-IQ2_XXS",
|
|
21876
|
+
slug: "unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21877
|
+
sourceURL: "https://huggingface.co/unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21878
|
+
vramTierGB: 128
|
|
21879
|
+
},
|
|
21880
|
+
{
|
|
21881
|
+
author: "unsloth",
|
|
21882
|
+
baseModel: "deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
21883
|
+
blurb: "The 4-bit cut of DeepSeek's experimental vision variant of V4-Flash - 284B MoE with 13B activated at higher fidelity, with multimodal agent scores well clear of the text-only release. Image input currently needs a preview llama.cpp build; needs 256 GB.",
|
|
21884
|
+
contextLength: 1048576,
|
|
21885
|
+
fileSizeBytes: 136662447706,
|
|
21886
|
+
format: "gguf",
|
|
21887
|
+
license: "mit",
|
|
21888
|
+
parameterCount: "284B",
|
|
21889
|
+
quantization: "UD-IQ4_XS",
|
|
21890
|
+
slug: "unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21891
|
+
sourceURL: "https://huggingface.co/unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
|
|
21892
|
+
vramTierGB: 256
|
|
21893
|
+
},
|
|
21698
21894
|
{
|
|
21699
21895
|
author: "unsloth",
|
|
21700
21896
|
baseModel: "Qwen/Qwen3.6-27B",
|
|
@@ -362930,7 +363126,7 @@ async function runModelFit(options) {
|
|
|
362930
363126
|
? Math.max(...detection.gpus
|
|
362931
363127
|
.map(gpu => gpu.memoryTotalBytes ?? 0)
|
|
362932
363128
|
.filter(bytes => bytes > 0))
|
|
362933
|
-
: detection.memory.totalBytes)}. Smallest tier starts at
|
|
363129
|
+
: detection.memory.totalBytes)}. Smallest tier starts at 2 GB.`);
|
|
362934
363130
|
process.exitCode = 1;
|
|
362935
363131
|
return;
|
|
362936
363132
|
}
|