Latent Node

[Latent Node @latent_node]

https://t.co/czP4jXQ9G3

[Latent Node @latent_node]

@suchenzang You need to treat writing as code and do it in claude code, it actually works well, You can plan ahead and create plotline, characters, tone etc. and you can ensure the agent adheres to them.

[Latent Node @latent_node]

@eastdakota This is more common than you think, had someone offer me to fire my co founder and only then they would invest.

[Latent Node @latent_node]

@ChanduThota if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model -https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@nickmhc @rasbt if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@IanBallantyne if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model - https://t.co/ZJPurl2hHd

[Latent Node @latent_node]

@googlegemma if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

@0xOzp @LottoLabs How did you get the MTP layer did oyu train it?

[Latent Node @latent_node]

@googlegemma if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@LyalinDotCom if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@KazimAIZJU if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@JustinLin610 if you use one of the optiq quants it is quite fast and fits in 16 gb VRAM amazing model - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@PaulGugAI if you use one of the optiq quants it is quite fast as well - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@matthewridenour @triswarkentin @o_lacombe @osanseviero @DynamicWebPaige That's wild it is also one of the most downlaoded optiq quants for mlx - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@JulianPscheid @HedyAI_ The model is really good for local agentic workflows and lightenting fast - https://t.co/MaGZBxzgO2

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

see - https://t.co/FDwNzWQZAF

[Latent Node @latent_node]

Gemma-4 12B, quantized with mlx-optiq, running on a Mac: 2.5x faster than bf16 (28.8 vs 11.6 tok/s on M3 Max), 2.7x smaller (8.9 GB vs 24 GB), fits in 16 GB of RAM. The sensitivity-aware 4-bit also beats naive 4-bit by +6.4 Capability Score (+13 long-context, +11.6
code, 93% GSM8K). No accuracy tax.

pip install mlx-optiq.

[Latent Node @latent_node]

@PaulGugAI @Teknium @NeoAIForecast The optiq quants tend to be more accurate and faster compared to the same uniform 4 bit ones on mlx. With MTP it will be quite fast - https://t.co/UH381cyvJt

[Latent Node @latent_node]

@ollama On Mac you can also try https://t.co/ewndTDNoMj for faster models

[Latent Node @latent_node]

@Scobleizer Yeah it does get flagged - https://t.co/zVdh1ljD4u

[Latent Node @latent_node]

@ChachraSahil Great work, saw the same model also recently used for a finetune - https://t.co/mFUM5TSukM