FINAL-Bench/POCKET-Image-Zimage
Text-to-Image ā¢ Updated ā¢ 24
Warecube Korea
Warecube
AI & ML interests
None yet
Recent Activity
liked a model 1 day ago
FINAL-Bench/POCKET-Image-Zimage liked a Space 1 day ago
FINAL-Bench/POCKET-Image-Studio reacted to SeaWolf-AI's post with š§ 2 days ago
POCKET now speaks Gemma 4 ā a 26B model that loads in every app, and runs on your PC with no GPU
We're adding a Gemma-4 sibling to POCKET: POCKET-26B, built from Google's Gemma-4-26B-A4B (Apache-2.0). Our flagship POCKET-35B is a Qwen-family MoE and needs a recent llama.cpp; POCKET-26B trades a little size for the thing people kept asking for ā it just loads, everywhere, today: Ollama, LM Studio, PocketPal, MLX, any stock llama.cpp. No fork, no bleeding-edge runtime, no CUDA, no cloud.
It's a sparse Mixture-of-Experts (25.2B total, ~4B active per token), so the work per token stays small ā a real 26B that generates on a CPU with no graphics card.
Two things make it stand out:
1) Universal compatibility. Gemma 4 is a standard, widely-supported architecture, so POCKET-26B runs on the tools you already have ā no waiting for your app to add a new model type.
2) Quality that survives compression. Measured GPQA-Diamond (198 q, greedy):
ā¢ Full base: 67.7%
ā¢ POCKET-26B Q4_K_M (17 GB): 67.7% ā lossless
ā¢ POCKET-26B Q2_K (11 GB): 67.2% ā near-lossless, at 11 GB
Live, on a CPU-only box (our demo Space ā POCKET-26B vs Bonsai-27B, same machine, same stock llama.cpp): POCKET-26B ā 19 tok/s vs Bonsai ā 6 tok/s ā about 3Ć faster generation, no GPU. (Honest notes: shared CPU box, sequential race; a dedicated machine is faster.)
Where it fits in the family:
ā¢ POCKET-35B (Qwen MoE) ā bigger, top-tier, needs a recent llama.cpp.
ā¢ POCKET-26B (Gemma 4) ā loads in any app, quality-robust when compressed. The demo runs the Q4_K_M build; Q2_K (11 GB) is the smallest footprint. For a true ā¤8 GB phone, the 5 GB POCKET-KR (Qwen) is still the pick.
Try it and grab it:
š„ļø Live demo (Gemma4-based, answering on a CPU, no GPU): https://huggingface.co/spaces/FINAL-Bench/POCKET-26B-CPU
š¦ POCKET-26B-GGUF (Q4_K_M 17 GB Ā· Q2_K 11 GB): https://huggingface.co/FINAL-Bench/POCKET-26B-GGUF
š POCKET collection: https://huggingface.co/collections/FINAL-Bench/pocket-modelsOrganizations
None yet
liked a
model 1 day ago
liked a
Space 1 day ago
reacted to SeaWolf-AI's post with š§ š„ 2 days ago
Post
2953
POCKET now speaks Gemma 4 ā a 26B model that loads in every app, and runs on your PC with no GPU
We're adding a Gemma-4 sibling to POCKET: POCKET-26B, built from Google's Gemma-4-26B-A4B (Apache-2.0). Our flagship POCKET-35B is a Qwen-family MoE and needs a recent llama.cpp; POCKET-26B trades a little size for the thing people kept asking for ā it just loads, everywhere, today: Ollama, LM Studio, PocketPal, MLX, any stock llama.cpp. No fork, no bleeding-edge runtime, no CUDA, no cloud.
It's a sparse Mixture-of-Experts (25.2B total, ~4B active per token), so the work per token stays small ā a real 26B that generates on a CPU with no graphics card.
Two things make it stand out:
1) Universal compatibility. Gemma 4 is a standard, widely-supported architecture, so POCKET-26B runs on the tools you already have ā no waiting for your app to add a new model type.
2) Quality that survives compression. Measured GPQA-Diamond (198 q, greedy):
ā¢ Full base: 67.7%
ā¢ POCKET-26B Q4_K_M (17 GB): 67.7% ā lossless
ā¢ POCKET-26B Q2_K (11 GB): 67.2% ā near-lossless, at 11 GB
Live, on a CPU-only box (our demo Space ā POCKET-26B vs Bonsai-27B, same machine, same stock llama.cpp): POCKET-26B ā 19 tok/s vs Bonsai ā 6 tok/s ā about 3Ć faster generation, no GPU. (Honest notes: shared CPU box, sequential race; a dedicated machine is faster.)
Where it fits in the family:
ā¢ POCKET-35B (Qwen MoE) ā bigger, top-tier, needs a recent llama.cpp.
ā¢ POCKET-26B (Gemma 4) ā loads in any app, quality-robust when compressed. The demo runs the Q4_K_M build; Q2_K (11 GB) is the smallest footprint. For a true ā¤8 GB phone, the 5 GB POCKET-KR (Qwen) is still the pick.
Try it and grab it:
š„ļø Live demo (Gemma4-based, answering on a CPU, no GPU): FINAL-Bench/POCKET-26B-CPU
š¦ POCKET-26B-GGUF (Q4_K_M 17 GB Ā· Q2_K 11 GB): FINAL-Bench/POCKET-26B-GGUF
š POCKET collection: https://huggingface.co/collections/FINAL-Bench/pocket-models
We're adding a Gemma-4 sibling to POCKET: POCKET-26B, built from Google's Gemma-4-26B-A4B (Apache-2.0). Our flagship POCKET-35B is a Qwen-family MoE and needs a recent llama.cpp; POCKET-26B trades a little size for the thing people kept asking for ā it just loads, everywhere, today: Ollama, LM Studio, PocketPal, MLX, any stock llama.cpp. No fork, no bleeding-edge runtime, no CUDA, no cloud.
It's a sparse Mixture-of-Experts (25.2B total, ~4B active per token), so the work per token stays small ā a real 26B that generates on a CPU with no graphics card.
Two things make it stand out:
1) Universal compatibility. Gemma 4 is a standard, widely-supported architecture, so POCKET-26B runs on the tools you already have ā no waiting for your app to add a new model type.
2) Quality that survives compression. Measured GPQA-Diamond (198 q, greedy):
ā¢ Full base: 67.7%
ā¢ POCKET-26B Q4_K_M (17 GB): 67.7% ā lossless
ā¢ POCKET-26B Q2_K (11 GB): 67.2% ā near-lossless, at 11 GB
Live, on a CPU-only box (our demo Space ā POCKET-26B vs Bonsai-27B, same machine, same stock llama.cpp): POCKET-26B ā 19 tok/s vs Bonsai ā 6 tok/s ā about 3Ć faster generation, no GPU. (Honest notes: shared CPU box, sequential race; a dedicated machine is faster.)
Where it fits in the family:
ā¢ POCKET-35B (Qwen MoE) ā bigger, top-tier, needs a recent llama.cpp.
ā¢ POCKET-26B (Gemma 4) ā loads in any app, quality-robust when compressed. The demo runs the Q4_K_M build; Q2_K (11 GB) is the smallest footprint. For a true ā¤8 GB phone, the 5 GB POCKET-KR (Qwen) is still the pick.
Try it and grab it:
š„ļø Live demo (Gemma4-based, answering on a CPU, no GPU): FINAL-Bench/POCKET-26B-CPU
š¦ POCKET-26B-GGUF (Q4_K_M 17 GB Ā· Q2_K 11 GB): FINAL-Bench/POCKET-26B-GGUF
š POCKET collection: https://huggingface.co/collections/FINAL-Bench/pocket-models
- 3 replies
liked a
Space 2 days ago
liked a
model 2 days ago
liked 9
models 3 days ago
mradermacher/Darwin-V9-Chimera-4B-i1-GGUF
4B ā¢ Updated ā¢ 883 ā¢ 3
mradermacher/Darwin-4B-Chimera-i1-GGUF
4B ā¢ Updated ā¢ 712 ā¢ 3
mradermacher/Darwin-V9-Chimera-4B-SFT-i1-GGUF
4B ā¢ Updated ā¢ 931 ā¢ 3
bartowski/FINAL-Bench_Darwin-4B-Chimera-GGUF
Text Generation ā¢ 4B ā¢ Updated ā¢ 1.12k ā¢ 3
Pq234/Darwin-28B-Opus
Text Generation ā¢ 28B ā¢ Updated ā¢ 353 ā¢ 4
mradermacher/Darwin-4B-Chimera-GGUF
4B ā¢ Updated ā¢ 386 ā¢ 5
usermma/Darwin-28B-Coder-mlx-fp16
Text Generation ā¢ 27B ā¢ Updated ā¢ 48 ā¢ 5
el4/Darwin-36B-Opus-OPAL-GGUF
Text Generation ā¢ 35B ā¢ Updated ā¢ 13.8k ā¢ 12
TheCluster/Darwin-31B-Opus-MLX-8bit
Image-Text-to-Text ā¢ 33B ā¢ Updated ā¢ 62 ā¢ 10
upvoted a paper 4 days ago
upvoted an article 4 days ago
Article
POCKET: a 35-billion-parameter model that runs on your iPhone ā and on your PC with no GPU
FINAL-Bench
ā¢ ā¢ 12 liked a
Space 4 days ago
liked 2
models 4 days ago