HardcoreAI/Nova-1.1-0.8B
Text Generation ā¢ 1B ā¢ Updated ā¢ 181 ā¢ 3
Rev Quick
Revertron
AI & ML interests
None yet
Recent Activity
liked a model 1 day ago
HardcoreAI/Nova-1.1-0.8B liked a model 5 days ago
mradermacher/SVGThinker-7B-GGUFOrganizations
None yet
liked a
model 1 day ago
liked a
model 5 days ago
reacted to wayneworkman2012's post with ā 6 days ago
Post
3,168
To all the inference providers behind the HuggingFace router - if you want me to use your services you will have to serve native-precision models. I'm not paying for FP8 when I can run in bf16 locally.
- 6 replies
reacted to SeaWolf-AI's post with š„ 6 days ago
Post
2,758
š We pretrained a language model from scratch on an inference-only NPU, with real backpropagation.
š¤ Model: FINAL-Bench/VKORE-46M-NPU š Technical write-up: https://huggingface.co/blog/FINAL-Bench/backpropagation-on-an-inference-only-npu-pretraini
"NPUs are inference-only, you can't train on them." We built a counterexample: VKORE-46M-NPU, trained from scratch on a first-gen (2025) inference NPU, 32 chips, no pretrained weights. A 46M decoder-only model (RoPE Ā· SwiGLU Ā· RMSNorm Ā· GQA), char-level, TinyStories.
š Why trust it: a falling-loss curve proves nothing. So we trained the same model under identical init, data, and schedule on the NPU and on a GPU reference. Ā· Loss trajectories correlate at 0.99 Ā· Throughput converged to a comparable level The NPU backward pass and optimizer updates do numerically the same work as the GPU, not an approximation that merely trends down.
š§© One accelerator now covers pretraining Ā· LoRA Ā· inference Ā· quantization. We call this layer VKORE: "CUDA for NPUs."
āļø Public: weights, model card, NPU-vs-GPU comparison. Private (patent-pending, paper to follow): the core training technique. 46M char-level is toy scale on purpose: it proves the training path, not model quality. Larger models and real data are next.
Feedback from folks working on custom autograd backends and low-precision training kernels on non-GPU accelerators is very welcome. š
š¤ Model: FINAL-Bench/VKORE-46M-NPU š Technical write-up: https://huggingface.co/blog/FINAL-Bench/backpropagation-on-an-inference-only-npu-pretraini
"NPUs are inference-only, you can't train on them." We built a counterexample: VKORE-46M-NPU, trained from scratch on a first-gen (2025) inference NPU, 32 chips, no pretrained weights. A 46M decoder-only model (RoPE Ā· SwiGLU Ā· RMSNorm Ā· GQA), char-level, TinyStories.
š Why trust it: a falling-loss curve proves nothing. So we trained the same model under identical init, data, and schedule on the NPU and on a GPU reference. Ā· Loss trajectories correlate at 0.99 Ā· Throughput converged to a comparable level The NPU backward pass and optimizer updates do numerically the same work as the GPU, not an approximation that merely trends down.
š§© One accelerator now covers pretraining Ā· LoRA Ā· inference Ā· quantization. We call this layer VKORE: "CUDA for NPUs."
āļø Public: weights, model card, NPU-vs-GPU comparison. Private (patent-pending, paper to follow): the core training technique. 46M char-level is toy scale on purpose: it proves the training path, not model quality. Larger models and real data are next.
Feedback from folks working on custom autograd backends and low-precision training kernels on non-GPU accelerators is very welcome. š
- 10 replies
liked a
model 9 days ago
liked a
model 10 days ago
I wish you delivered what you say you delivered
šš 5
16
#3 opened about 2 months ago
by 
NezTheNaughty
reacted to mihailgribov's post with š¤Æ 14 days ago
Post
2,310
TypeCastLM: Jev-class decision models on frozen LLMs
TypeCastLM is a decision model built around a frozen LLM. Given a text and a question, it returns calibrated probabilities over the allowed answers in one forward pass. The LLM is not fine-tuned: not one of its weights changes. TypeCastLM swaps its LM head for a mini one: a linear matrix whose rows are mostly the model's own output rows, with the three verdict rows (true, false, unsure) fitted. A fitted row catches an answer the model spreads over many tokens, such as "yes", "true" or "correct", where a single vocabulary row sees one word. A head that small leaves little room to overfit. It speaks the Jev API, so a Jev client switches by changing the base URL.
The first model, [typecastlm-qwen3.5-3.8b]( mihailgribov/typecastlm-qwen3.5-3.8b) on Qwen3.5-4B, is among the top open frozen-4B models on JevBench v1.6. Besides Jev's yes/no, choice and scale modes it has
Specs:
ā¢ Size: 3.76B params; 7.5 GB bf16, 4.0 GB Q8_0
ā¢ VRAM: 12 GB is enough, 16 GB comfortable
ā¢ Latency p50 (batch 1, bf16, RTX 5060 Ti): ā¤200 tok 48 ms Ā· 1k 93 ms Ā· 4k 566 ms
ā¢ Calibration (ECE): BoolQ 0.052 Ā· RTE 0.012 Ā· FEVER 0.019
ā¢ Context: 32k tokens
ā¢ Runtime: transformers or llama.cpp, fully offline
ā¢ JevBench v1.6: 19.7
TypeCastLM is a decision model built around a frozen LLM. Given a text and a question, it returns calibrated probabilities over the allowed answers in one forward pass. The LLM is not fine-tuned: not one of its weights changes. TypeCastLM swaps its LM head for a mini one: a linear matrix whose rows are mostly the model's own output rows, with the three verdict rows (true, false, unsure) fitted. A fitted row catches an answer the model spreads over many tokens, such as "yes", "true" or "correct", where a single vocabulary row sees one word. A head that small leaves little room to overfit. It speaks the Jev API, so a Jev client switches by changing the base URL.
The first model, [typecastlm-qwen3.5-3.8b]( mihailgribov/typecastlm-qwen3.5-3.8b) on Qwen3.5-4B, is among the top open frozen-4B models on JevBench v1.6. Besides Jev's yes/no, choice and scale modes it has
tfu: yes/no with a third answer, unsure, for when the text lacks what the decision needs.Specs:
ā¢ Size: 3.76B params; 7.5 GB bf16, 4.0 GB Q8_0
ā¢ VRAM: 12 GB is enough, 16 GB comfortable
ā¢ Latency p50 (batch 1, bf16, RTX 5060 Ti): ā¤200 tok 48 ms Ā· 1k 93 ms Ā· 4k 566 ms
ā¢ Calibration (ECE): BoolQ 0.052 Ā· RTE 0.012 Ā· FEVER 0.019
ā¢ Context: 32k tokens
ā¢ Runtime: transformers or llama.cpp, fully offline
ā¢ JevBench v1.6: 19.7
pip install typecastlm 
- 2 replies
liked a
model 15 days ago
liked 2
models 16 days ago
liked a
dataset 17 days ago
Great goal! I'm also trying something like that, but more as an SFT for fine-tuning. But now I've pivoted to a sophisticated library inside my coding-agent...
reacted to cafkafk's post with š„ 20 days ago
Post
4,496
I'm working on a local model that can be run on stuff like an an RTX 3080 at 60 t/s. I'm focused on making it good at Rust and Nix. It's... not done yet, but doing great already. Had a 28/100 on livebench v6 earlier, which obviously isn't amazing.
But for my first real model, very exciting. Can't wait to share more as I get closer to the finish line in the coming weeks.
But for my first real model, very exciting. Can't wait to share more as I get closer to the finish line in the coming weeks.
- 6 replies
liked a
dataset 21 days ago
liked 2
models 22 days ago
liked a
dataset 23 days ago
It can't use tools?
#1 opened 23 days ago
by Revertron