@andrewfarah This worked a treat, but I have a question: How would I output the classification ran by the LLM. The processed data/classification seems accessible by ft, partly visible as a nice summary with ft vis, but I can’t seem to find the output myself. Is it meant to be temporary?
@mrbleu_11 @typedfemale@grok They look like uperfect delta monitors, probably 2x the max 18”, which are unwieldy to travel with IMO. It has HDMI in and usb c in for each screen on each dual monitor. You can’t feed 2 signals from one comp over a single usb c cable. You should avoid monitors that promise that
@bilawalsidhu Is there a proper simple 360 to splat workflow yet, or is the process still breaking a spherical image into rectilinear/flat images to process?
@Minttt876@HuggingModels@grok Via Gemini:
VRAM Requirements (GPU) for the 4-bit quantized version (GPTQ-Int4)
• Model Weights: ~68–72 GB
KV Cache (Context): additional overhead for the "memory" of the conversation.
• Short context (8k tokens): ~75 GB VRAM.
• Full context (256k tokens): >100 GB+
@grok
@DarioWB1@HuggingModels@grok Via Gemini:
VRAM Requirements (GPU) for the 4-bit quantized version (GPTQ-Int4)
• Model Weights: ~68–72 GB
KV Cache (Context): additional overhead for the "memory" of the conversation.
• Short context (8k tokens): ~75 GB VRAM.
• Full context (256k tokens): >100 GB+
@HuggingModels Via Gemini:
VRAM Requirements (GPU) for the 4-bit quantized version (GPTQ-Int4)
• Model Weights: ~68–72 GB
KV Cache (Context): additional overhead for the "memory" of the conversation.
• Short context (8k tokens): ~75 GB VRAM.
• Full context (256k tokens): >100 GB+