@TencentHunyuan There's a gap in the 128GB market. A 2bit quant could become a leader in that segment. Any chance of getting Q2 and Q3 quants to experiment with?
@TheAhmadOsman An 80-90B dense would dominate the market at 128GB along with their 27B. Such a huge opportunity to fill a market gap being wasted. I hope they prove me wrong.
@julianharris@iamtrask Great high-level read! I’ll be trying omlx and I’m glad you link to examples. You might want to also mention that the output quality of these local models is lower than cloud/frontier but will improve in your predictions section so readers don’t assume they’re at parity today.
@julianharris@iamtrask I’m in the same boat. So many model runners, quants, tools (turboquant, dflash), etc. it’s hard not to get trapped into a paralysis of choice. Good luck!
@julianharris@iamtrask Which Gemma model/quant are you using? I’m setting up a new claw with same system but it’s hard finding the local “sweet spot” model to get the most reliable responses/harness utilization.
@ale1735187@TheAhmadOsman Which 8bit KV quant versions are you using for these two or did you roll your own? I’d like to give them a try for myself if they’re on HF.
Happening Now: Group walking around handing out envelopes with $200 cash to convoy participants. This was on Wellington an hour ago. @OttawaPolice does this activity count as illegally "assisting others in the blocking streets"? #Ottawa#ottnews
Happening Now: Group walking around handing out envelopes with $200 cash to convoy participants. This was on Wellington an hour ago. @OttawaPolice does this activity count as illegally "assisting others in the blocking streets"? #Ottawa#ottnews