Also what is Klein?
DedeProGames PRO
DedeProGames
AI & ML interests
Agentic & Coding finetune, Decoder-Onlys from Scratch, Image LoRAs, AI Researcher
Recent Activity
liked a model about 4 hours ago
LiquidAI/LFM3-1.7B new activity 1 day ago
OrionLLM/GRM-3.2-Sky:GRM Sky based on Ornith 1.5 or Tiel? new activity 1 day ago
OrionLLM/GRM-3.2-Sky:Correct benchmarks - GRM-2.6-Plus-0628 to GRM-2.6-PlusOrganizations
replied to Banaxi-Tech's post 3 days ago
replied to Banaxi-Tech's post 3 days ago
@Banaxi-Tech You should get ready, because SupraLabs is cooking with Supra2.5. 🔥🔥🔥
reacted to Banaxi-Tech's post with 🚀🔥 3 days ago
Post
5,545
BananaMind 3 is going to be an comeback. Sorry for the BGA things ( i genuinely didnt know about nsa existing and yes I used aiiii)
Were going to use our BananaMind 2.1 Unified architecture with some changes + NSA and some other stuff.
Were going to use our BananaMind 2.1 Unified architecture with some changes + NSA and some other stuff.
- 40 replies
reacted to Banaxi-Tech's post with 😔 4 days ago
Post
4,874
BananaMind is in the google situation before argon right now, right?
- 29 replies
reacted to nightmedia's post with 😔 5 days ago
Post
2,542
The DavidAU/Qwen3.6-27B-Fable-Fusion-711
This model was assembled at NightmediaAI under the lab name:
nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451
There is ample information on the model card about the model itself.
Here are the base model metrics, as it was merged
It is a merge of different models, some abliterated, some as they were, so as a result, this source is not Heretic. DavidAU offers the source in Heretic mode here:
DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP
The basic idea behind this model was to build a model that can be used as a base, and has high enough IQ to merge with other models. Every model card shows details of assembly and provides a vibe check with the Nightmedia test prompts.
I merged a few samples here:
with nbeerbower/BigBubba-Qwen3.6-27B
nightmedia/Qwen3.6-27B-Fable-Fusion-711-BigBubba-717-mxfp8-mlx
with migtissera/Tess-4-27B:
nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Tess-mxfp8-mlx
with EpistemeAI/Reasoning-Medical-27B (metrics pending):
nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Medical-mxfp8-mlx
This last model was simply a blind merge, and might not be the proper formula for it, but was entertaining enough to publish as a sample.
I am populating the model card metrics as they become available.
This model was assembled at NightmediaAI under the lab name:
nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451
There is ample information on the model card about the model itself.
Here are the base model metrics, as it was merged
Qwen3.6-27B-Architect-Polaris2-Fable-B-F451
arc arc/e boolq hswag obkqa piqa wino
mxfp8 0.711,0.879,0.910,0.790,0.514,0.823,0.763
qx86-hi 0.696,0.876,0.912,0.791,0.518,0.824,0.760
qx64-hi 0.702,0.873,0.909,0.794,0.514,0.822,0.750
mxfp4 0.701,0.873,0.909,0.786,0.488,0.813,0.759It is a merge of different models, some abliterated, some as they were, so as a result, this source is not Heretic. DavidAU offers the source in Heretic mode here:
DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP
The basic idea behind this model was to build a model that can be used as a base, and has high enough IQ to merge with other models. Every model card shows details of assembly and provides a vibe check with the Nightmedia test prompts.
I merged a few samples here:
with nbeerbower/BigBubba-Qwen3.6-27B
nightmedia/Qwen3.6-27B-Fable-Fusion-711-BigBubba-717-mxfp8-mlx
mxfp8 0.717,0.873,0.912
mxfp4 0.707,0.877,0.912with migtissera/Tess-4-27B:
nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Tess-mxfp8-mlx
mxfp8 0.712,0.879,0.911,0.792,0.508,0.823,0.764
mxfp4 0.706,0.873,0.910,0.790,0.496,0.817,0.761with EpistemeAI/Reasoning-Medical-27B (metrics pending):
nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Medical-mxfp8-mlx
mxfp8 0.716,0.877,0.911
mxfp4 0.701This last model was simply a blind merge, and might not be the proper formula for it, but was entertaining enough to publish as a sample.
I am populating the model card metrics as they become available.
- 6 replies
replied to their post 8 days ago
Thanks for the suggestion! I ran the CPU benchmark and Portuguese-IFEval. The F16 GGUF scored 6.54% strict prompt accuracy and decoded at 65.5 tokens/s with llama.cpp, versus 10.8 tokens/s in PyTorch FP32. The chat template passed all 9 checks. I couldn’t finish the Q4_K_M quality run, so I don’t have a valid Q4 score yet.
replied to their post 10 days ago
Ent, e dificil ahcar aqui kkkkkk
replied to their post 12 days ago
Im from Brazil
reacted to their post with 🤯🧠🚀🔥 12 days ago
Post
8,902
🚀 Introducing Kiyo-230M-Instruct — my compact English & Portuguese chat model!
Built on Kiyo-230M-Preview, a base model I pretrained from scratch, Kiyo now has an instruction-tuned release for everyday conversations, rewriting, and lightweight assistant experiments.
🧠 229.7M parameters
🌍 English + Portuguese
💬 Native chat template for multi-turn conversations
📏 2,048-token context window
📦 459 MB of BF16 safetensors weights
🔓 Apache-2.0 license
For instruction tuning, I used 165,735 conversations from SmolTalk2, smol-smoltalk, UltraChat, and Dolci-Instruct-SFT. Training ran for two complete epochs with assistant-only loss, using a roughly 70% English / 30% Portuguese token mix.
The updated model card includes benchmarks against SmolLM2-135M-Instruct, BananaMind, and Supra2, plus example conversations, CPU measurements, training details, and a Transformers quickstart.
The results are mixed: Kiyo is competitive with these small models across the reported evaluations, while SmolLM2 and BananaMind score higher on the instruction-following benchmarks. Portuguese was a highlight in the small conversational sanity test.
This is an experimental 230M model, with plenty of room to improve reasoning, factual accuracy, and strict instruction following. I’m excited to share it with people exploring how much we can get from small, accessible models!
🤗 Model & full README:
DedeProGames/Kiyo-230M-Instruct
💬 Try the chat demo:
DedeProGames/kiyo-230m-chat
Try it in English or Portuguese — I’d love to see your examples, feedback, and experiments!
Built on Kiyo-230M-Preview, a base model I pretrained from scratch, Kiyo now has an instruction-tuned release for everyday conversations, rewriting, and lightweight assistant experiments.
🧠 229.7M parameters
🌍 English + Portuguese
💬 Native chat template for multi-turn conversations
📏 2,048-token context window
📦 459 MB of BF16 safetensors weights
🔓 Apache-2.0 license
For instruction tuning, I used 165,735 conversations from SmolTalk2, smol-smoltalk, UltraChat, and Dolci-Instruct-SFT. Training ran for two complete epochs with assistant-only loss, using a roughly 70% English / 30% Portuguese token mix.
The updated model card includes benchmarks against SmolLM2-135M-Instruct, BananaMind, and Supra2, plus example conversations, CPU measurements, training details, and a Transformers quickstart.
The results are mixed: Kiyo is competitive with these small models across the reported evaluations, while SmolLM2 and BananaMind score higher on the instruction-following benchmarks. Portuguese was a highlight in the small conversational sanity test.
This is an experimental 230M model, with plenty of room to improve reasoning, factual accuracy, and strict instruction following. I’m excited to share it with people exploring how much we can get from small, accessible models!
🤗 Model & full README:
DedeProGames/Kiyo-230M-Instruct
💬 Try the chat demo:
DedeProGames/kiyo-230m-chat
Try it in English or Portuguese — I’d love to see your examples, feedback, and experiments!
- 9 replies
posted an update 12 days ago
Post
8,902
🚀 Introducing Kiyo-230M-Instruct — my compact English & Portuguese chat model!
Built on Kiyo-230M-Preview, a base model I pretrained from scratch, Kiyo now has an instruction-tuned release for everyday conversations, rewriting, and lightweight assistant experiments.
🧠 229.7M parameters
🌍 English + Portuguese
💬 Native chat template for multi-turn conversations
📏 2,048-token context window
📦 459 MB of BF16 safetensors weights
🔓 Apache-2.0 license
For instruction tuning, I used 165,735 conversations from SmolTalk2, smol-smoltalk, UltraChat, and Dolci-Instruct-SFT. Training ran for two complete epochs with assistant-only loss, using a roughly 70% English / 30% Portuguese token mix.
The updated model card includes benchmarks against SmolLM2-135M-Instruct, BananaMind, and Supra2, plus example conversations, CPU measurements, training details, and a Transformers quickstart.
The results are mixed: Kiyo is competitive with these small models across the reported evaluations, while SmolLM2 and BananaMind score higher on the instruction-following benchmarks. Portuguese was a highlight in the small conversational sanity test.
This is an experimental 230M model, with plenty of room to improve reasoning, factual accuracy, and strict instruction following. I’m excited to share it with people exploring how much we can get from small, accessible models!
🤗 Model & full README:
DedeProGames/Kiyo-230M-Instruct
💬 Try the chat demo:
DedeProGames/kiyo-230m-chat
Try it in English or Portuguese — I’d love to see your examples, feedback, and experiments!
Built on Kiyo-230M-Preview, a base model I pretrained from scratch, Kiyo now has an instruction-tuned release for everyday conversations, rewriting, and lightweight assistant experiments.
🧠 229.7M parameters
🌍 English + Portuguese
💬 Native chat template for multi-turn conversations
📏 2,048-token context window
📦 459 MB of BF16 safetensors weights
🔓 Apache-2.0 license
For instruction tuning, I used 165,735 conversations from SmolTalk2, smol-smoltalk, UltraChat, and Dolci-Instruct-SFT. Training ran for two complete epochs with assistant-only loss, using a roughly 70% English / 30% Portuguese token mix.
The updated model card includes benchmarks against SmolLM2-135M-Instruct, BananaMind, and Supra2, plus example conversations, CPU measurements, training details, and a Transformers quickstart.
The results are mixed: Kiyo is competitive with these small models across the reported evaluations, while SmolLM2 and BananaMind score higher on the instruction-following benchmarks. Portuguese was a highlight in the small conversational sanity test.
This is an experimental 230M model, with plenty of room to improve reasoning, factual accuracy, and strict instruction following. I’m excited to share it with people exploring how much we can get from small, accessible models!
🤗 Model & full README:
DedeProGames/Kiyo-230M-Instruct
💬 Try the chat demo:
DedeProGames/kiyo-230m-chat
Try it in English or Portuguese — I’d love to see your examples, feedback, and experiments!
- 9 replies
replied to Banaxi-Tech's post 13 days ago
@Banaxi-Tech please take a look at this
https://huggingface.co/DedeProGames/Kiyo-230M-Instruct
reacted to Banaxi-Tech's post with 🚀 13 days ago
Post
3,813
KV1 updates:
We at
klein-ml have finished stage 1 of KV1 1.2B, we don't think we will need additional stages.
KV1 2.6B is near the midpoint of training, going well so far, its going to be our most agentic model yet.
We have released Klein Code Bench klein-ml/klein-code-bench, a benchmark testing whether models can build complete projects from scratch. It contains a easy, intermediate, hard and max subset.
Here are some benchmarks of the current KV1 1.2B
It retains 75.8% of FP16 intelligence.
We are also announcing that we at klein-ml are working on Klein 24B A2B which will be our first model featuring ternary weight quantization.
We at
KV1 2.6B is near the midpoint of training, going well so far, its going to be our most agentic model yet.
We have released Klein Code Bench klein-ml/klein-code-bench, a benchmark testing whether models can build complete projects from scratch. It contains a easy, intermediate, hard and max subset.
Here are some benchmarks of the current KV1 1.2B
It retains 75.8% of FP16 intelligence.
We are also announcing that we at
- 2 replies
replied to Banaxi-Tech's post 13 days ago
Gotta bencgmark OxCoder and GRM on this 🔥
reacted to aryan-sabi's post with 🤯 13 days ago
Post
3,799
tl;dr: new qwen finetune that simultaneously pushes the frontiers of small model capability and reasoning efficiency.
Hey everyone,
We are Sabi (https://sabi.com), a neuro-AI startup building a BCI + AI wearable that can do agentic actions using just your neural intents.
We publish two post-trained versions of Qwen 3.8 27B called Sabi-Noema and Sabi-Noema-MLA. These models push the capability frontier on <30B models with 2x reduction in verbosity, and are exclusively trained on open source environments.
These models have improved capabilities on coding, agentic and tool calling tasks, as well as hallucinating lesser. Sabi-Noema beats GPT 6 Luna-medium on DeepSWE, placing in between Luna 6 medium and high. It beats Luna-high on agentic and tool calling benchmarks like TB2.1, Tau-3 bench, ClawEval, and BFCL v4. Its hallucination rate is considerably lower than Luna, Qwen and other competing models.
These models are much more reasoning efficient. Sabi-Noema generates 50% fewer tokens per response on TB2.1 and 25% fewer tokens on DeepSWE, compared to base Qwen. Sabi-Noema-MLA goes even further by doing a post-hoc MLA conversion of Qwen's GQA attention. This conversion uses only a calibration dataset and no extra training. With BF16 latents Sabi-Noema-MLA needs 3.6x lower bytes per token. The MLA latents can be further compressed to FP8 with the total KV cache size on TB2.1 to 8.4x smaller with similar performance. To put this in perspective, when using Sabi-Noema-MLA at q4_0, on a 24GB GPU like 3090/4090 you can get a context size of ~534K where base Qwen at FP8 cache would only get to ~150K context.
We are excited to have the open source community play around with these models. We will be releasing more post-trains in the future focused on making <30B models more powerful.
HuggingFace:
Sabi-Noema: sabicap/sabi-noema
Sabi-Noema-MLA: sabicap/sabi-noema-mla
Hey everyone,
We are Sabi (https://sabi.com), a neuro-AI startup building a BCI + AI wearable that can do agentic actions using just your neural intents.
We publish two post-trained versions of Qwen 3.8 27B called Sabi-Noema and Sabi-Noema-MLA. These models push the capability frontier on <30B models with 2x reduction in verbosity, and are exclusively trained on open source environments.
These models have improved capabilities on coding, agentic and tool calling tasks, as well as hallucinating lesser. Sabi-Noema beats GPT 6 Luna-medium on DeepSWE, placing in between Luna 6 medium and high. It beats Luna-high on agentic and tool calling benchmarks like TB2.1, Tau-3 bench, ClawEval, and BFCL v4. Its hallucination rate is considerably lower than Luna, Qwen and other competing models.
These models are much more reasoning efficient. Sabi-Noema generates 50% fewer tokens per response on TB2.1 and 25% fewer tokens on DeepSWE, compared to base Qwen. Sabi-Noema-MLA goes even further by doing a post-hoc MLA conversion of Qwen's GQA attention. This conversion uses only a calibration dataset and no extra training. With BF16 latents Sabi-Noema-MLA needs 3.6x lower bytes per token. The MLA latents can be further compressed to FP8 with the total KV cache size on TB2.1 to 8.4x smaller with similar performance. To put this in perspective, when using Sabi-Noema-MLA at q4_0, on a 24GB GPU like 3090/4090 you can get a context size of ~534K where base Qwen at FP8 cache would only get to ~150K context.
We are excited to have the open source community play around with these models. We will be releasing more post-trains in the future focused on making <30B models more powerful.
HuggingFace:
Sabi-Noema: sabicap/sabi-noema
Sabi-Noema-MLA: sabicap/sabi-noema-mla
- 7 replies
reacted to their post with 🧠🚀 14 days ago
Post
3,459
Meet Supra2.5-Ultra-Preview-001. 230M parameters, ~40.4B training tokens.
SupraLabs has released the first preview of its Supra2.5-Ultra family: a compact, Llama-style base model for text completion and small-model experimentation.
What's inside?
• 230M parameters
• 32 transformer layers
• 2,048-token context window
• Grouped Query Attention: 12 attention heads, 4 KV heads
• 49,152-token vocabulary
• SwiGLU activation, RoPE and tied embeddings
• BF16 weights and Hugging Face Transformers support
• Apache 2.0 license
Training
The model was trained on approximately 40.4B tokens from FineWeb-Edu, DCLM-Baseline and FineMath using a single A100 SXM.
Reported zero-shot results
• HellaSwag: 37.18% normalized accuracy
• PIQA: 65.78% accuracy
• ARC-Easy: 57.11% accuracy
• ARC-Challenge: 26.28% normalized accuracy
Want to experiment?
Code on REPO.
Model, weights and usage example:
SupraLabs/Supra2.5-Ultra-Preview-001
SupraLabs has released the first preview of its Supra2.5-Ultra family: a compact, Llama-style base model for text completion and small-model experimentation.
What's inside?
• 230M parameters
• 32 transformer layers
• 2,048-token context window
• Grouped Query Attention: 12 attention heads, 4 KV heads
• 49,152-token vocabulary
• SwiGLU activation, RoPE and tied embeddings
• BF16 weights and Hugging Face Transformers support
• Apache 2.0 license
Training
The model was trained on approximately 40.4B tokens from FineWeb-Edu, DCLM-Baseline and FineMath using a single A100 SXM.
Reported zero-shot results
• HellaSwag: 37.18% normalized accuracy
• PIQA: 65.78% accuracy
• ARC-Easy: 57.11% accuracy
• ARC-Challenge: 26.28% normalized accuracy
Want to experiment?
Code on REPO.
Model, weights and usage example:
SupraLabs/Supra2.5-Ultra-Preview-001
- 4 replies