dmw_ng

12 ms·

91.
by dmw_ng 4y ago
Big fan of ClickHouse, but in the case of S3 there doesn't seem to be much point fighting city hall. I was the author of the original clickhouse-local-in-Lambda ticket FWIW. Even if that approach worked well, it'd still only be &l
92.
by dmw_ng 4y ago
If someone were following the setup steps to execute a query in a serious setting (as opposed to reviewing them just for the sake of writing a sarcastic forum comment), they might find CREATE TABLE .. MSCK REPAIR TABLE quickly becomes finge
93.
by dmw_ng 4y ago
Casually querying an S3 bucket with either of those is easily a half day's work for any meaningful size of workload. S3 Select and Athena make this very straightforward, and in the case of Athena+parquet, obscenely cheap given the capa
94.
by dmw_ng 4y ago
The LoRa weights for 13B are on huggingface ( https://huggingface.co/samwit/alpaca13B-lora ), it should be possible to follow the instructions linked in the Alpaca.cpp readme to merge those weights into the base 13B mo
95.
Alpaca training code and data released(twitter.com)
4 points by dmw_ng 4y ago|0 comments
96.
by dmw_ng 4y ago
The only thing Google "released" was an announcement of a waiting list, that as of a few hours ago twitter folk claimed didn't even exist yet
97.
by dmw_ng 4y ago
It's adorable seeing this kind of critique in the context of HN, I wonder how many folk knew my heavily upvoted comment above ("For 10 billion+ ...") was from someone who has only been looking at this stuff for a few weeks. C
98.
by dmw_ng 4y ago
There is some very natural split regarding what I'll send to ChatGPT vs. what goes to Google. For example "six nations fixtures" obviously Google, but anything of depth or where recency is irrelevant goes the other direction.
99.
by dmw_ng 4y ago
For 10 billion+ parameter models, the effects of quantization are relatively small, for smaller models like Llama 7B the effect becomes more dramatic, but there is ongoing research on new quantization methods (like GPTQ) that preserve signi
100.
by dmw_ng 4y ago
I also noticed hitting CTRL+S to pause the TTY output seemed to cause a reliable prompt to suddenly start printing garbage tokens after CTRL+Q to resume a few seconds later. It may have been a coincidence, but instant thought was very much
101.
by dmw_ng 4y ago
just wanted to say thanks for your effort in aggregating and communicating a fast moving and extremely interesting area, have been watching your output like a hawk recently
102.
by dmw_ng 4y ago
with llama.cpp, you might need to pass in -t to set the thread count. What kind of OS / host environment are you using? I noticed very little speedup with using t=16 and t=32, it's possible the code simply hasn't been tested
103.
by dmw_ng 4y ago
Can anyone recommend a UI clone as close to the regular ChatGPT as possible, but built on the API? The API traffic is sent directly into Azure while the regular UI is sent via CloudFlare. Vastly prefer the former reading my conversations th
104.
by dmw_ng 4y ago
Both, the API response includes a breakdown. In the best case 1 token = 1 word (for example "and", "the", etc). Depending on input, for English it seems reasonable to multiply the word count by about 1.3 to get a rough t
105.
by dmw_ng 4y ago
It may just be that OpenAI corrected its 2020-era understanding of model size using 2022 insights from DeepMind.. https://towardsdatascience.com/a-new-ai-trend-chinchilla-70b... Seems model sizing, compression and quantizat
106.
by dmw_ng 4y ago
I have been using it as a search replacement for most of the past month and only found two subtly wrong answers. This covers legal questions, researching product differences, wiring diagrams, suggesting books to read, correcting misremember
107.
by dmw_ng 4y ago
https://www.cambridge.org/core/journals/perspectives-on-poli...
108.
by dmw_ng 4y ago
"Google uses Bayesian filtering the way Microsoft uses the if statement"* This quote is 18 years old, and already by then it was established understanding that Google had extreme technical magic the rest of the world lacked. A rus
109.
by dmw_ng 4y ago
Just to add the cost-per-query may not be a temporary blip. ChatGPT's cost per query is based on a static base model trained over a year ago. Imagine the energy and compute that might be required to add meaningful personalization or an
110.
by dmw_ng 4y ago
https://www.youtube.com/watch?v=npV4Kix7Td0 Responses on Twitter are horrific: https://twitter.com/googleeurope/status/1623251768099409920 This is the first non-technical Google presentation I&#x2
111.
by dmw_ng 4y ago
I'd like to try out your service, but would not be willing to link my GitHub account, and I do not have a Google account. Any chance of a non-federated sign in method?
112.
by dmw_ng 4y ago
Curious where you heard the detail about Amazon
113.
by dmw_ng 4y ago
ITAR according to Twitter https://twitter.com/vk5qi/status/1591628725141270528 https://spectrum.ieee.org/passive-radar-with-sdr
114.
by dmw_ng 4y ago
Ideally it's only necessary transmit about 1 frame per minute for each half-open connection, but you're right, I found one site, India's rail operator IRCTC that still blocked this traffic very quickly.
115.
by dmw_ng 4y ago
The original problem was for load balancing, where the protocol is often not known, and backends might be reached over a WAN. There are some neat advantages possible from this method, such as the target application not being woken up early
116.
by dmw_ng 4y ago
This can also be used to implement connection acceleration with no kernel-side hackery involved, by having userspace maintain a pool of half-open connections by spamming SYNs at a target, recording ACKs, then constructing a regular socket v
117.
by dmw_ng 4y ago
As an uneducated farmer this gave me a giggle, it suggests if I filed a ticket about a performance problem in the project it'd likely be fixed instantaneously. That quote is basically an advert, if the author wishes to establish such a
118.
by dmw_ng 4y ago
Dorsey is an engineer
119.
by dmw_ng 4y ago
They definitely had permanent data loss in App Engine around a decade ago, along with leaving any app active during the outage with copies of every database table to manually merge
120.
by dmw_ng 4y ago
This seems unfair, it isn't necessary to attack the author just because the tone is a little conspiratorial. I'm sure there is a little embellishment in the post, but there doesn't seem to be any reason to doubt the author ha

More ›