ProCreations/grug-9b
Text Generation • 9B • Updated • 488 • 6
John Smith PRO
John6666
AI & ML interests
None yet
Recent Activity
liked a model about 1 hour ago
ProCreations/grug-9b liked a model about 1 hour ago
ProCreations/grug-9b-ggufOrganizations
liked 2
models about 1 hour ago
reacted to ProCreations's post with 🔥 about 1 hour ago
Post
906
grug model here. archives 10 percent worse performance but with a 3-4 times speed bost because grug good. try now.
ProCreations/grug-9b
ProCreations/grug-9b-gguf
ProCreations/grug-9b
ProCreations/grug-9b-gguf
reacted to satgeze's post with 🔥 about 1 hour ago
Post
139
satgeze/Ornith-1.0-35B-1M-GGUF
Ornith-1.0 with a 1,048,576-token context window, tested instead of claimed 🦜
Ornith is Qwen3.5-family under the hood, so YaRN factor 4 extends it from 262K native to exactly 1M. I baked that into the GGUF metadata (no fine-tuning, weights bit-identical) so llama.cpp and Ollama apply it with zero flags, then ran full needle-in-a-haystack ladders on my own hardware:
- satgeze/Ornith-1.0-35B-1M-GGUF: 10/10 needles at every rung from 32K through 1M, replicated with fresh seeds (M3 Max 128GB, ~6.8h cold 1M prefill)
- satgeze/Ornith-1.0-9B-1M-GGUF: perfect through 524K, honest 7/10 at 1M under Q4 + q8_0 KV, failure band charted in the card
- satgeze/Ornith-1.0-397B-1M-GGUF: IQ1_M through Q4_K_M as split GGUFs, coherence-gated
Also in the repos:
- Vision: Ornith kept the Qwen3.5 multimodal skeleton, so the VL vision tower (extracted by bartowski) attaches at runtime via llama-server --mmproj. OCR-tested on the 9B and 35B, mmproj files bundled.
- A measured residency matrix: on a single RTX 5090, every 9B quant up to Q6_K holds the full 1M window at 100 percent GPU, 162 to 244 tok/s.
- Quality gates: every low-bit quant passed a coherence test before upload. The 35B IQ1_S failed and was deleted rather than shipped.
Harness, method writeup, and raw per-needle data: https://github.com/satindergrewal/ornith-1m
All MIT. Credit to DeepReinforce for the models and bartowski for the imatrix quants and vision towers. If a config breaks retrieval for you, tell me and it goes in the card.
Ornith-1.0 with a 1,048,576-token context window, tested instead of claimed 🦜
Ornith is Qwen3.5-family under the hood, so YaRN factor 4 extends it from 262K native to exactly 1M. I baked that into the GGUF metadata (no fine-tuning, weights bit-identical) so llama.cpp and Ollama apply it with zero flags, then ran full needle-in-a-haystack ladders on my own hardware:
- satgeze/Ornith-1.0-35B-1M-GGUF: 10/10 needles at every rung from 32K through 1M, replicated with fresh seeds (M3 Max 128GB, ~6.8h cold 1M prefill)
- satgeze/Ornith-1.0-9B-1M-GGUF: perfect through 524K, honest 7/10 at 1M under Q4 + q8_0 KV, failure band charted in the card
- satgeze/Ornith-1.0-397B-1M-GGUF: IQ1_M through Q4_K_M as split GGUFs, coherence-gated
Also in the repos:
- Vision: Ornith kept the Qwen3.5 multimodal skeleton, so the VL vision tower (extracted by bartowski) attaches at runtime via llama-server --mmproj. OCR-tested on the 9B and 35B, mmproj files bundled.
- A measured residency matrix: on a single RTX 5090, every 9B quant up to Q6_K holds the full 1M window at 100 percent GPU, 162 to 244 tok/s.
- Quality gates: every low-bit quant passed a coherence test before upload. The 35B IQ1_S failed and was deleted rather than shipped.
Harness, method writeup, and raw per-needle data: https://github.com/satindergrewal/ornith-1m
All MIT. Credit to DeepReinforce for the models and bartowski for the imatrix quants and vision towers. If a config breaks retrieval for you, tell me and it goes in the card.
- 2 replies
upvoted an article about 1 hour ago
Article
Quantum Cryptanalysis on Real Hardware: Pushing Symmetric-Structure Key Recovery Beyond the Published Frontier
FINAL-Bench
• • 15 reacted to SeaWolf-AI's post with 🔥 about 1 hour ago
Post
4969
🔓 We ran genuine quantum key-recovery on 'real IBM quantum hardware' — and pushed the frontier well past the largest hardware demos we're aware of (which sat at N=4).
Using Simon's algorithm on
• Even–Mansour — N=5 → N=10
• 3-round Feistel (DES-family) — block 6 → 8
Each verified against an 'independent control key', using error mitigation only (no QEC).
🧭 Honest scope: this is not a quantum speedup (the effective difficulty tracks the classical birthday bound ~2^{n/2}), not a break of real AES/RSA, and not 16-round DES (ours is 3-round). The recovery method is reserved for a forthcoming paper; formal record status is pending peer review.
📄 Write-up: https://huggingface.co/blog/FINAL-Bench/quantum
🕹️ Try it live in your browser: https://vidraft-quantumos.hf.space/crypto
🏆 Leaderboard: FINAL-Bench/quantum-bench-leaderboard
#quantum #cryptography #quantumcomputing
Using Simon's algorithm on
ibm_kingston, we recovered the secret key of two symmetric-cipher structures:• Even–Mansour — N=5 → N=10
• 3-round Feistel (DES-family) — block 6 → 8
Each verified against an 'independent control key', using error mitigation only (no QEC).
🧭 Honest scope: this is not a quantum speedup (the effective difficulty tracks the classical birthday bound ~2^{n/2}), not a break of real AES/RSA, and not 16-round DES (ours is 3-round). The recovery method is reserved for a forthcoming paper; formal record status is pending peer review.
📄 Write-up: https://huggingface.co/blog/FINAL-Bench/quantum
🕹️ Try it live in your browser: https://vidraft-quantumos.hf.space/crypto
🏆 Leaderboard: FINAL-Bench/quantum-bench-leaderboard
#quantum #cryptography #quantumcomputing
reacted to pranavupadhyaya52's post with 🔥 about 1 hour ago
Post
3053
After several weeks of experimenting, debugging, and iterating, I am excited to share WikiSmartBotLM.
WikiSmartBotLM is a compact decoder only language model built from the ground up as an educational and practical project. The goal was not simply to train another language model, but to create one that is easy to understand, modify, and experiment with while following many of the architectural ideas used in modern LLMs.
The model is built on a custom Transformer architecture featuring Rotary Positional Embeddings, RMSNorm, SwiGLU feed forward layers, grouped query attention, and an efficient autoregressive decoder optimized for local inference.
The repository includes the complete model implementation, configuration files, tokenizer integration, training pipeline, inference scripts, checkpoint conversion utilities, and examples that demonstrate how each component works together. Whether you want to understand the forward pass, train your own model, or build applications on top of WikiSmartBotLM, everything is designed to be approachable.
You can directly run the model via the models Huggingfaces space, which I've included in the post.
Model Repository: pranavupadhyaya52/Wiki-SmartBotLM-Instruct
I hope WikiSmartBotLM becomes a useful resource for anyone who enjoys learning by building. Feedback, issues, feature requests, and contributions are always welcome.
https://huggingface.co/spaces/pranavupadhyaya52/WikiSmartBot
WikiSmartBotLM is a compact decoder only language model built from the ground up as an educational and practical project. The goal was not simply to train another language model, but to create one that is easy to understand, modify, and experiment with while following many of the architectural ideas used in modern LLMs.
The model is built on a custom Transformer architecture featuring Rotary Positional Embeddings, RMSNorm, SwiGLU feed forward layers, grouped query attention, and an efficient autoregressive decoder optimized for local inference.
The repository includes the complete model implementation, configuration files, tokenizer integration, training pipeline, inference scripts, checkpoint conversion utilities, and examples that demonstrate how each component works together. Whether you want to understand the forward pass, train your own model, or build applications on top of WikiSmartBotLM, everything is designed to be approachable.
You can directly run the model via the models Huggingfaces space, which I've included in the post.
Model Repository: pranavupadhyaya52/Wiki-SmartBotLM-Instruct
I hope WikiSmartBotLM becomes a useful resource for anyone who enjoys learning by building. Feedback, issues, feature requests, and contributions are always welcome.
https://huggingface.co/spaces/pranavupadhyaya52/WikiSmartBot
- 1 reply
reacted to Bc-AI's post with 👍 about 1 hour ago
Post
88
New Update on Nova-1 series status!!! So, after 3 days of fixing our dataset script, we finally have nova-1-standard in its final phases of instruction tuning hopefully. I genuinely do not know ha-ha. We are also doing a novel model named Nova-1-EXP with 5 novel components in the model, which we will announce when the time comes.
reacted to cetusian's post with 👀 about 1 hour ago
Post
85
most labs won't bother teaching a small model to write and think in real romanian. no invented words, no english leaking in mid-sentence, diacritics intact.
we did.
surogate 3.5 is out. 2B and 4B, apache 2.0. invented word-forms cut from 4 per 1k down to ~1.5, and it reasons in the language you prompt it in, every time.
intelligence should speak your language too.
the models:
surogate/Surogate-3.5-4B
surogate/Surogate-3.5-2B
we did.
surogate 3.5 is out. 2B and 4B, apache 2.0. invented word-forms cut from 4 per 1k down to ~1.5, and it reasons in the language you prompt it in, every time.
intelligence should speak your language too.
the models:
surogate/Surogate-3.5-4B
surogate/Surogate-3.5-2B
reacted to a9090's post with 👀 about 1 hour ago
Post
66
Building in Public: How KALE.ng is Fighting AI-Slacking in Education
I just shipped a major performance fix on KALE.ng — our platform that aims to support faculty to transform educational assessment from "catch AI text" to "prove human reasoning over time." - The Knowledge Authentication Layer for Education.
More importantly, I just put together both a long form draft of a case study of the changes made after the first 70 days and a short form of the case (linked below). While building this platform, I'm equally interested in drawing out practical case studies that present students with the opportunity to engage with the fine points of AI enabled application development.
KALE.ng is live. Educators, try the assignment analyzer. Students, read through the brief case and see if it sparks some thoughts on how you may better design your next App.
Go build! Build, in the name of progress!
https://kale.ng/research/kale-launch-case-study/
I just shipped a major performance fix on KALE.ng — our platform that aims to support faculty to transform educational assessment from "catch AI text" to "prove human reasoning over time." - The Knowledge Authentication Layer for Education.
More importantly, I just put together both a long form draft of a case study of the changes made after the first 70 days and a short form of the case (linked below). While building this platform, I'm equally interested in drawing out practical case studies that present students with the opportunity to engage with the fine points of AI enabled application development.
KALE.ng is live. Educators, try the assignment analyzer. Students, read through the brief case and see if it sparks some thoughts on how you may better design your next App.
Go build! Build, in the name of progress!
https://kale.ng/research/kale-launch-case-study/
reacted to ovi054's post with 🚀 about 1 hour ago
Post
4216
Turn any topic to animation explainer video
It is powered by Qwen3-14B + Manim LoRA and turns any concept into a Manim explainer video.
👉 Try it now:
build-small-hackathon/anim-vid-ai
It is powered by Qwen3-14B + Manim LoRA and turns any concept into a Manim explainer video.
👉 Try it now:
build-small-hackathon/anim-vid-ai
reacted to Banaxi-Tech's post with 🔥 about 1 hour ago
Post
4149
Today we are releasing BananaMind-KV1-8M-2Bit-Experimental, a KV-cache-aware trained model that stores its generation KV cache in 2-bit precision instead of the usual 16-bit precision.
Result: 5.33x smaller KV cache vs FP16, with 0.0916 mean KLD against a 16-bit KV cache reference on WikiText-2.
Model: BananaMind/BananaMind-KV1-8M-2Bit-Experimental
The important part: this is not just post-training KV cache quantization.
Instead we take the BitNet approach.
KV1 is trained with a 2-bit-aware K/V path. Instead of training a normal model and quantizing the cache afterwards, the model learns during training to operate under the low-bit KV constraint, closer in spirit to the BitNet idea of training for the low-bit regime.
During generation, each K/V vector is quantized into 4 affine levels and packed into uint8 tensors, with four 2-bit values stored per byte.
WikiText-2 eval vs 16-bit KV cache reference:
Mean KLD: 0.0916 nats/token
Mean KLD: 0.1322 bits/token
Average KV cache shrink vs FP16: 5.33x
Evaluated positions: 372,675
If this actually gets used in models like Qwen or Gemma, then it may be possible to run 128K or even 256K Context on a Normal Machine!
Try it here: BananaMind/BananaMind-KV1-8M-2Bit-Experimental
Code: https://github.com/Banaxi-Tech/kv1
Result: 5.33x smaller KV cache vs FP16, with 0.0916 mean KLD against a 16-bit KV cache reference on WikiText-2.
Model: BananaMind/BananaMind-KV1-8M-2Bit-Experimental
The important part: this is not just post-training KV cache quantization.
Instead we take the BitNet approach.
KV1 is trained with a 2-bit-aware K/V path. Instead of training a normal model and quantizing the cache afterwards, the model learns during training to operate under the low-bit KV constraint, closer in spirit to the BitNet idea of training for the low-bit regime.
During generation, each K/V vector is quantized into 4 affine levels and packed into uint8 tensors, with four 2-bit values stored per byte.
WikiText-2 eval vs 16-bit KV cache reference:
Mean KLD: 0.0916 nats/token
Mean KLD: 0.1322 bits/token
Average KV cache shrink vs FP16: 5.33x
Evaluated positions: 372,675
If this actually gets used in models like Qwen or Gemma, then it may be possible to run 128K or even 256K Context on a Normal Machine!
Try it here: BananaMind/BananaMind-KV1-8M-2Bit-Experimental
Code: https://github.com/Banaxi-Tech/kv1
- 4 replies
reacted to pankajpandey-dev's post with 🚀 about 1 hour ago
Post
4030
🇮🇳 Qwen3.5-9B Hindi Instruct — it stops thinking in English
Ask base Qwen3.5-9B a question in Hindi and it burns hundreds of tokens thinking in English inside its think block before a single Devanagari word appears — then code-switches in the answer. I fine-tuned it to close the think block instantly and reply in pure, native Hindi.
✅ Model (16-bit): pankajpandey-dev/qwen3.5-9b-hindi-instruct
✅ GGUF (Q4/Q5/Q8): pankajpandey-dev/qwen3.5-9b-hindi-instruct-GGUF
✅ Try it in the browser: pankajpandey-dev/qwen3.5-9b-hindi-demo
Recipe: Unsloth + LoRA (r=16, response-only loss) on 12.9k Hindi pairs — AI4Bharat anudesh + dolly-hi + wikiHow-hi + Aya Hindi (human-written). The Q4_K_M is 5.4 GB and runs on a plain laptop CPU.
New in this run vs my earlier models: mixed in long-form native sources (wikiHow) after my last eval showed the fine-tune traded detail for conciseness — this one keeps answers detailed and native.
Part of my weekly 🇮🇳 Hindi LLM Series. Feedback welcome 🙏
#Hindi #IndicNLP #Qwen #GGUF #LocalLLM #Unsloth
Ask base Qwen3.5-9B a question in Hindi and it burns hundreds of tokens thinking in English inside its think block before a single Devanagari word appears — then code-switches in the answer. I fine-tuned it to close the think block instantly and reply in pure, native Hindi.
✅ Model (16-bit): pankajpandey-dev/qwen3.5-9b-hindi-instruct
✅ GGUF (Q4/Q5/Q8): pankajpandey-dev/qwen3.5-9b-hindi-instruct-GGUF
✅ Try it in the browser: pankajpandey-dev/qwen3.5-9b-hindi-demo
Recipe: Unsloth + LoRA (r=16, response-only loss) on 12.9k Hindi pairs — AI4Bharat anudesh + dolly-hi + wikiHow-hi + Aya Hindi (human-written). The Q4_K_M is 5.4 GB and runs on a plain laptop CPU.
New in this run vs my earlier models: mixed in long-form native sources (wikiHow) after my last eval showed the fine-tune traded detail for conciseness — this one keeps answers detailed and native.
Part of my weekly 🇮🇳 Hindi LLM Series. Feedback welcome 🙏
#Hindi #IndicNLP #Qwen #GGUF #LocalLLM #Unsloth
- 4 replies
reacted to Quazim0t0's post with 👍 about 1 hour ago
Post
87
I built a small Neural Physics Engine from scratch
The idea I wanted to test: in a projective-dynamics solver, could you replace each hand-derived local constraint projection with a learned one, while keeping the analytic parts (rotations, the global solve) exactly as they are? It's one tiny network, shared across every element and across constraint types through material tokens. A new material isn't a new network, just a new token row. Fluids fall out of the same idea, with water treated as one more token.
A few things held up in testing: one tied projector matched five separate per-material solutions, the neural fluid tracked the exact analytic solver closely on a dam-break sim, and a learned warm-start trimmed solver iterations without touching correctness.
Try it here:
Dam Break Demo:
https://quazim0t0-neural-physics-engine-demo.hf.space/
FPS Shooter Demo:
https://quazim0t0-neural-combat-evolved.hf.space/
Model Repo:
Quazim0t0/neural-physics-engine
The idea I wanted to test: in a projective-dynamics solver, could you replace each hand-derived local constraint projection with a learned one, while keeping the analytic parts (rotations, the global solve) exactly as they are? It's one tiny network, shared across every element and across constraint types through material tokens. A new material isn't a new network, just a new token row. Fluids fall out of the same idea, with water treated as one more token.
A few things held up in testing: one tied projector matched five separate per-material solutions, the neural fluid tracked the exact analytic solver closely on a dam-break sim, and a learned warm-start trimmed solver iterations without touching correctness.
Try it here:
Dam Break Demo:
https://quazim0t0-neural-physics-engine-demo.hf.space/
FPS Shooter Demo:
https://quazim0t0-neural-combat-evolved.hf.space/
Model Repo:
Quazim0t0/neural-physics-engine
reacted to kanaria007's post with 👀 about 1 hour ago
Post
61
✅ Article highlight: Treaty Objects: Partial Agreement, Reserved Disagreement, and Bounded Cross-System Commit (art-60-191, v0.1)
TL;DR:
This article argues that “the parties agreed” is not enough.
Governed systems often need joint action without full trust, shared values, or perfect equivalence. 191 defines Treaty Objects: bounded cross-party commitments that preserve reserved disagreement, bind only the shared commit surface, split shared/local execution, and include expiry, appeal, exit, rollback, and reentry semantics.
Read:
kanaria007/agi-structural-intelligence-protocols
Why it matters:
• enables partial agreement without pretending full alignment
• keeps disagreement explicit instead of laundering it into consensus
• prevents a treaty from becoming the union of all local authority
• separates treaty-bound shared execution from local-only actions
• makes expiry, appeal, exit, suspension, and reentry part of the artifact
What’s inside:
• treaty scope for included and excluded surfaces
• reserved-disagreement manifests with per-clause impact
• commit clauses with required refs, preconditions, effect class, and rollback mode
• dual-channel execution records for shared and local channels
• expiry, appeal, and exit clauses
• compression-aware and epoch-aware treaty behavior
• activation, suspension, appeal, exit, closure, and reentry receipts
Key idea:
Do not say:
“the institutions aligned.”
Say:
“these parties bound this limited treaty scope, preserved these disagreements, activated only these clauses, split shared and local execution, and left lifecycle receipts for appeal, exit, expiry, rollback, and reentry.”
Treaties are bounded promises.
Not dishonest unity.
TL;DR:
This article argues that “the parties agreed” is not enough.
Governed systems often need joint action without full trust, shared values, or perfect equivalence. 191 defines Treaty Objects: bounded cross-party commitments that preserve reserved disagreement, bind only the shared commit surface, split shared/local execution, and include expiry, appeal, exit, rollback, and reentry semantics.
Read:
kanaria007/agi-structural-intelligence-protocols
Why it matters:
• enables partial agreement without pretending full alignment
• keeps disagreement explicit instead of laundering it into consensus
• prevents a treaty from becoming the union of all local authority
• separates treaty-bound shared execution from local-only actions
• makes expiry, appeal, exit, suspension, and reentry part of the artifact
What’s inside:
• treaty scope for included and excluded surfaces
• reserved-disagreement manifests with per-clause impact
• commit clauses with required refs, preconditions, effect class, and rollback mode
• dual-channel execution records for shared and local channels
• expiry, appeal, and exit clauses
• compression-aware and epoch-aware treaty behavior
• activation, suspension, appeal, exit, closure, and reentry receipts
Key idea:
Do not say:
“the institutions aligned.”
Say:
“these parties bound this limited treaty scope, preserved these disagreements, activated only these clauses, split shared and local execution, and left lifecycle receipts for appeal, exit, expiry, rollback, and reentry.”
Treaties are bounded promises.
Not dishonest unity.
reacted to ProCreations's post with 🧠 about 1 hour ago
Post
70
grug model soon from grug dataset. grug fine tune ornith. grug life.
upvoted a collection about 1 hour ago
reacted to juiceb0xc0de's post with 🔥 about 1 hour ago
Post
80
HF community! Thank you so much the overwhelming flood of downloads I've received on my model atlases over the last couple of weeks. It makes me so happy to know other people out there are enjoying my work. With that being said, I am looking for an opportunity from the builders and trainers out there.
I've recently had the chance to do a mid pre training snapshot durin process and am looking for the opportunity to get some more of these images . This is a symbiotic trade. I am more than happy to do full reports on the status of your model and translate what your model may stand to gain from as you head towards the finish line. The atlas building process uses 8965 different prompt over a wide variety of behavioural features. I combines many different methods of interpreting what is going on inside the model between prompt and response that we are unable to see.
For anyone else interested in neural network imaging and mechanistic interpretation check out my library of atlases and my new work translating the numbers into a 3d visual format to accompany the sqlite database. If you've gpt a model that you wish to have atlased leave a commen and I'kk be sure to ge arounf to it as soon as possible.
https://huggingface.co/collections/juiceb0xc0de/cloud-atlases
I've recently had the chance to do a mid pre training snapshot durin process and am looking for the opportunity to get some more of these images . This is a symbiotic trade. I am more than happy to do full reports on the status of your model and translate what your model may stand to gain from as you head towards the finish line. The atlas building process uses 8965 different prompt over a wide variety of behavioural features. I combines many different methods of interpreting what is going on inside the model between prompt and response that we are unable to see.
For anyone else interested in neural network imaging and mechanistic interpretation check out my library of atlases and my new work translating the numbers into a 3d visual format to accompany the sqlite database. If you've gpt a model that you wish to have atlased leave a commen and I'kk be sure to ge arounf to it as soon as possible.
https://huggingface.co/collections/juiceb0xc0de/cloud-atlases