He said/implied that we would have shortage either way (without data center buildout, but everyone doing private/local inference), which doesn't sound far fetched.
> and the remainder of VRAM is occupied by the KV cache and whatever experts still fit. Those days engines try to optimize "whatever experts" part and can use remaining VRAM as cache, moving hot experts to GPU during…
There's good open weight models for now, labs could stop releasing them for whatever reason (assuming "good model" is a moving target).
Moved where? Moving to a different wallet wouldn't prevent freezing.
Luna/Terra/Sol/Astra make perfect sense, it indicates model tier/size with larger celestial body representing larger model. Same for Anthropic, consistent model tier naming reused across model generations. Those are…
For whatever map rendering in Google Maps is capped at 30fps, yes. Google doesn't seem to care.
Deepseek doesn't have subscriptions, you can only pay per token / API prices.
Maybe it's intentional so government doesn't block open weight release. There was law proposal (rumour?) last week about Chinese labs having to get approved for weights release now.
There's basic reference inference code released with the weights, yes. https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017/... > providers have to implement which could be different than what the labs themselves…
From "What about data and privacy?" in FAQ it sounds like they don't host models themselves. e.g. it says: DeepSeek: ZDR agreement is renewed monthly. The current agreement is valid through October 31, 2026.…
Is that a dig on DeepSeek not offering subscription plans? If so OpenCode Go looks good and includes DeepSeek models (yet to try myself).
If you're happy with generation speed in 5 t/s range, sure -- PCIe 5.0 x16 is only 64GB/s.
158GB number is for Mistral Medium 3.5 128B model, Muse Spark is way larger. Muse Spark Open would require more than 1.6TB of RAM in its native precision (BF16) for the model parameters alone. In a 4-bit quantized…
For one it checks integrity of all resources for auto update & client repair purposes on each start. You could break/remove any file from install folder, it'll fix it. That might sound silly, but likely really useful…
I guess US `tabled` means put in a writing table drawer, hidden and soon to forgotten.
I'm reading schleck8 comment as accusing Yandex ASN as malicious on grounds of having higher bot % traffic than Scaleway / comparably sized Cloud Provider. It could be malicious, but that exact stat doesn't show that.
Where you saw whataboutism? Implication was that tons of malicious traffic originate from there. I pointed out that bot % stat doesn't seem to be useful for determining that, Occam's razor and all.
Scaleway doesn't have a search engine? Google ASN is at 98.7% https://radar.cloudflare.com/traffic/as15169
First key on the right after space bar.
There's no BF16, original full quality weights are quantized already and 510GB. Then good portion of those weights are n-grams (~200GB) that don't need to be in VRAM. Then KV cache of that model is super lightweight at…
I guess public datasets on HuggingFace and some shadow libraries content is enough to start. e.g. fineweb dataset is 50TB https://huggingface.co/datasets/HuggingFaceFW/fineweb
6B activated weights per token vs 27B. Something like DGX Spark is way better suited for Flash Next.
That's max output tokens per response limit, separate from context length
Devil's advocate, but Anthropic talks a lot about importance of alignment, so.. How can military ensure that model is NOT refusing to work just to be malicious? Anthropic doesn't provide model w/o guardrails, suppliers…
n-grams can be kept on SSD, no need to hold them in any kind of RAM (at least w/o batching)
He said/implied that we would have shortage either way (without data center buildout, but everyone doing private/local inference), which doesn't sound far fetched.
> and the remainder of VRAM is occupied by the KV cache and whatever experts still fit. Those days engines try to optimize "whatever experts" part and can use remaining VRAM as cache, moving hot experts to GPU during…
There's good open weight models for now, labs could stop releasing them for whatever reason (assuming "good model" is a moving target).
Moved where? Moving to a different wallet wouldn't prevent freezing.
Luna/Terra/Sol/Astra make perfect sense, it indicates model tier/size with larger celestial body representing larger model. Same for Anthropic, consistent model tier naming reused across model generations. Those are…
For whatever map rendering in Google Maps is capped at 30fps, yes. Google doesn't seem to care.
Deepseek doesn't have subscriptions, you can only pay per token / API prices.
Maybe it's intentional so government doesn't block open weight release. There was law proposal (rumour?) last week about Chinese labs having to get approved for weights release now.
There's basic reference inference code released with the weights, yes. https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017/... > providers have to implement which could be different than what the labs themselves…
From "What about data and privacy?" in FAQ it sounds like they don't host models themselves. e.g. it says: DeepSeek: ZDR agreement is renewed monthly. The current agreement is valid through October 31, 2026.…
Is that a dig on DeepSeek not offering subscription plans? If so OpenCode Go looks good and includes DeepSeek models (yet to try myself).
If you're happy with generation speed in 5 t/s range, sure -- PCIe 5.0 x16 is only 64GB/s.
158GB number is for Mistral Medium 3.5 128B model, Muse Spark is way larger. Muse Spark Open would require more than 1.6TB of RAM in its native precision (BF16) for the model parameters alone. In a 4-bit quantized…
For one it checks integrity of all resources for auto update & client repair purposes on each start. You could break/remove any file from install folder, it'll fix it. That might sound silly, but likely really useful…
I guess US `tabled` means put in a writing table drawer, hidden and soon to forgotten.
I'm reading schleck8 comment as accusing Yandex ASN as malicious on grounds of having higher bot % traffic than Scaleway / comparably sized Cloud Provider. It could be malicious, but that exact stat doesn't show that.
Where you saw whataboutism? Implication was that tons of malicious traffic originate from there. I pointed out that bot % stat doesn't seem to be useful for determining that, Occam's razor and all.
Scaleway doesn't have a search engine? Google ASN is at 98.7% https://radar.cloudflare.com/traffic/as15169
First key on the right after space bar.
There's no BF16, original full quality weights are quantized already and 510GB. Then good portion of those weights are n-grams (~200GB) that don't need to be in VRAM. Then KV cache of that model is super lightweight at…
I guess public datasets on HuggingFace and some shadow libraries content is enough to start. e.g. fineweb dataset is 50TB https://huggingface.co/datasets/HuggingFaceFW/fineweb
6B activated weights per token vs 27B. Something like DGX Spark is way better suited for Flash Next.
That's max output tokens per response limit, separate from context length
Devil's advocate, but Anthropic talks a lot about importance of alignment, so.. How can military ensure that model is NOT refusing to work just to be malicious? Anthropic doesn't provide model w/o guardrails, suppliers…
n-grams can be kept on SSD, no need to hold them in any kind of RAM (at least w/o batching)