Merged
Conversation
Contributor
|
Yes, testing with Qwen3.5-35B-A3B MoE 
👈 Detailsmain@216f4436 default batches (-ub 512 -b 2048)model=/models/ubergarm/Qwen3.5-35B-A3B-GGUF/Qwen3.5-35B-A3B-Q4_0.gguf
./build/bin/llama-sweep-bench \
--model "$model" \
-c 98816 \
--n-predict 128 \
-ger \
--merge-qkv \
-ngl 99 \
--threads 1 \
--warmup-batch
main@216f4436 -ub 2048 -b 2048model=/models/ubergarm/Qwen3.5-35B-A3B-GGUF/Qwen3.5-35B-A3B-Q4_0.gguf
./build/bin/llama-sweep-bench \
--model "$model" \
-c 36864 \
--n-predict 128 \
-ger \
--merge-qkv \
-ngl 99 \
-ub 2048 -b 2048 \
--threads 1 \
--warmup-batch
PR1320 ik/fused_delta_net_2@0579a868 -ub 2048 -b 2048 -fdn 512./build/bin/llama-sweep-bench \
--model "$model" \
-c 98816 \
--n-predict 128 \
-fdn 512 \
-ger \
--merge-qkv \
-ngl 99 \
--threads 1 \
--warmup-batch
PR1330 ik/slightly_better_fdn@c115e185 -ub 2048 -b 2048 -fdn 512./build/bin/llama-sweep-bench \
--model "$model" \
-c 98816 \
--n-predict 128 \
-fdn 512 \
-ger \
--merge-qkv \
-ngl 99 \
--threads 1 \
--warmup-batch
|
Owner
Author
|
So, on CUDA the fused delta-net is still slightly slower than the chunked implementation for batch sizes greater than 64 or so. Hence, if running on CUDA, it is best to use On the CPU (at least on my CPU) the fused delta net is faster than the chunked implementation, so there one can use I have been trying to make the fused delta-net be as fast as the chunked implementation for all batch sizes, but no luck so far. If I managed to get there, it would simplify things a lot! |
|
Okay cool its working. +~1% decode. Qwen3.5 IQ2_KL, 8x3090: Detailsmain: n_kv_max = 262144, n_batch = 4096, n_ubatch = 4096, flash_attn = 1, n_gpu_layers = 99, n_threads = 64, n_threads_batch = 64
refs: |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
~1% or less performance gain with full GPU offload.