Fanrong Li 4632a8642d [None][doc] blog: Optimizing DeepSeek-V3.2 on NVIDIA Blackwell GPUs (#10565) ... Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>		2026-01-09 05:16:00 -05:00
..
Falcon180B-H200_acc.png
Falcon180B-H200_DecvOct.png
Falcon180B-H200_H200vA100.png
Falcon180B-H200_tps.png
H200launch_H200vsH100_tps.png
H200launch_tps.png
moe_structure.png
tech_blog1_fuse_a_gemm.png
tech_blog1_model_details.png
tech_blog1_model_overview.png
tech_blog1_router_gemm.png
tech_blog1_sparse_exp_as_a_gemm.png
tech_blog2_acc_relaxed_acceptance.png
tech_blog2_mtp_eagle.png
tech_blog2_mtp_modules.png
tech_blog2_mtp_vanilla.png
tech_blog2_overall_workflow.png
tech_blog2_perf_and_ar.png
tech_blog2_relaxed_acceptance.png
tech_blog2_tree_spec_decoding.png
tech_blog2_verify_and_accept.png
tech_blog3_mla_absorb.png
tech_blog4_Picture1.png
tech_blog4_Picture2.png
tech_blog4_Picture3.png
tech_blog4_Picture4.png
tech_blog4_Picture5.png
tech_blog4_Picture6.png
tech_blog4_Picture7.png
tech_blog4_Picture8.png
tech_blog4_Picture9.png
tech_blog4_Picture10.png
tech_blog4_Picture11.png
tech_blog4_Picture12.png
tech_blog4_Picture13.png
tech_blog4_Picture14.png
tech_blog4_Picture15.png
tech_blog4_Picture16.png
tech_blog4_Picture17.png
tech_blog4_Picture18.png
tech_blog4_Picture19.png
tech_blog4_Picture20.png
tech_blog4_Picture21.png
tech_blog4_Picture22.png
tech_blog4_Picture23.png
tech_blog4_Picture24.png
tech_blog4_Picture25.png
tech_blog5_Picture1.png
tech_blog5_Picture2.png
tech_blog5_Picture3.png
tech_blog5_Picture4.png
tech_blog5_Picture5.png
tech_blog5_Picture6.png
tech_blog5_Picture7.png
tech_blog5_Picture8.png
tech_blog5_Picture9.png
tech_blog5_Picture10.png
tech_blog5_Picture11.png
tech_blog5_Picture12.png
tech_blog5_Picture13.png
tech_blog5_Picture14.png
tech_blog5_Picture15.png
tech_blog7_accepted_length_case2.png
tech_blog7_al_over_iteration_magpie.png
tech_blog7_init_sequence_scan.png
tech_blog7_magpie_accepted_length_distribution.png
tech_blog7_per_token_update.png
tech_blog7_speed_up_first_turn.png
tech_blog7_speed_up_second_turn.png
tech_blog8_communication_kernel.png
tech_blog8_kernel_breakdown.png
tech_blog8_moe_aux_kernels1.png
tech_blog8_moe_aux_kernels2.png
tech_blog8_perf-1k-1k-dep.png
tech_blog8_perf-4k-1k-dep.png
tech_blog8_perf-8k-1k-dep.png
tech_blog8_perf-8k-1k-e2e-mtp.png
tech_blog10_baseline_performance_detail.png
tech_blog10_baseline_performance_overview.png
tech_blog10_baseline_round_robin_strategy.png
tech_blog10_context_wait_performance.png
tech_blog10_dataset_token_distribution.png
tech_blog10_full_strategy_performance.png
tech_blog10_tps_ttft_pareto_curve.png
tech_blog12_constrained_decoding_pipeline_overlap.png
tech_blog12_cpu_gpu_synchronization_for_multiple_steps_by_cuda_callback.png
tech_blog12_cpu_gpu_synchronization_for_multiple_steps.png
tech_blog12_one_model_vs_two_model.png
tech_blog12_pareto_curve_json_mode_eval_llama_3.1_8b.png
tech_blog12_pareto_curve_json_mode_eval_llama_3.3_70b.png
tech_blog12_pareto_curve_json_schema_bench_llama_3.1_8b.png
tech_blog12_pareto_curve_json_schema_bench_llama_3.3_70b.png
tech_blog13_dynasor_demo.gif
tech_blog13_dynasor_hesitation.png
tech_blog13_dynasor_illustration.jpg
tech_blog13_dynasor_pressure_testing.png
tech_blog13_scaffolding_sequence.png
tech_blog14_alltoall_dataflow.png
tech_blog14_MTP_parallel_1.png
tech_blog14_MTP_parallel_2.png
tech_blog14_overview_after_opt.png
tech_blog14_overview_before_opt.png
tech_blog14_pdloff.png
tech_blog14_pdlon.png
tech_blog14_perf.png
tech_blog15_ds32_wide_ep.png	[None][doc] blog: Optimizing DeepSeek-V3.2 on NVIDIA Blackwell GPUs (#10565)	2026-01-09 05:16:00 -05:00
tech_blog15_dsa_architecture.png	[None][doc] blog: Optimizing DeepSeek-V3.2 on NVIDIA Blackwell GPUs (#10565)	2026-01-09 05:16:00 -05:00
tech_blog15_indexer_topk.png	[None][doc] blog: Optimizing DeepSeek-V3.2 on NVIDIA Blackwell GPUs (#10565)	2026-01-09 05:16:00 -05:00
tech_blog15_radix_select_topk.png	[None][doc] blog: Optimizing DeepSeek-V3.2 on NVIDIA Blackwell GPUs (#10565)	2026-01-09 05:16:00 -05:00
tp_ep.png
TRT_LLM_v0-5-0_H100vA100_1st.png
TRT_LLM_v0-5-0_H100vA100_tps.png
XQA_ThroughputvsLatency.png