ManniX-ITA/Qwen3.5-27B-Omnimerge-v2
28B β’ Updated β’ 18 β’ 1
ManniX PRO
ManniX-ITA
AI & ML interests
None yet
Recent Activity
updated a model about 2 hours ago
ManniX-ITA/Qwen3.6-27B-Omnimerge-v4 updated a model about 2 hours ago
ManniX-ITA/Qwen3.6-27B-Omnimerge-v4-GGUF posted an update about 10 hours ago
π Gemma-4-A4B 98e v5-coder β code-leaning 20.8B MoE (4B-active), C6 layer-relevance-weighted prune of Gemma 4 26B-A4B. Best 20B-class coder I've shipped.
π SCORES (NVFP4A16, vLLM 0.20.2, greedy, EVAL_PROTOCOL v3)
HumanEval 98.17 β HumanEval+ 92.68 β LCB-medium-55 v4 85.45
MATH-500 92.00 β GPQA-D 68.69 β IFEval 94.00
vs v4: +1.22 HE / +1.22 HE+ / +7.27 LCB-medium
Top of the 14β22B coder band: +8.6pp HE over Qwen2.5-Coder-14B-Instruct (89.6 β 98.17). HE+ sanity-audited β no memorization, no silent-empty.
π¦ EXTENSIVE GGUF SWEEP (16 plain + IQ tiers + 5 CD recipes, all imatrix-calibrated)
Q8_0 β 21.16 GB β 93.90% (cohort top)
Q4_K_S β 12.21 GB β 93.29% β plain sweet spot
IQ4_XS β 11.01 GB β 93.29% β sub-12 GB top
β TWO EXCELLENT SUB-10 GB CONTRIBDYNAMIC CD PICKS (per-layer + IQ-codebook overrides)
CD-IQ4_K_M (Canary W) β 10.29 GB β 92.07% β recommended sub-11 GB
CD-IQ3_XS_L β 9.27 GB β 90.24% β smallest viable code-grade
βοΈ SAME-RIG vs Qwen2.5-Coder-14B-Instruct (RTX 3090, greedy HE+)
11 GB band: v5-coder IQ4_XS wins +9.75pp at -1.49 bpw
12 GB band: Q4_K_S wins +8.53pp
8 GB band: IQ2_S wins +0.61pp at lower bpw
bf16:
https://huggingface.co/ManniX-ITA/gemma-4-A4B-98e-v5-coder-it
GGUF:
https://huggingface.co/ManniX-ITA/gemma-4-A4B-98e-v5-coder-it-GGUF
NVFP4A16:
https://huggingface.co/ManniX-ITA/gemma-4-A4B-98e-v5-coder-NVFP4A16
Ollama:
https://ollama.com/mannix/gemma4-98e-v5-coder
βββ
π BONUS β Qwen3.6-27B-Omnimerge-v4-MTP-GGUF
Same v4 weights with the native MTP head retained for llama.cpp speculative decoding (PR #22673, --spec-type draft-mtp). 7 imatrix tiers Q8_0 β IQ2_M.
HumanEval: 2.0x decode tok/s
MBPP: 2.33x decode tok/s
Both at +1-2pp pass@1 vs the non-MTP build. GPQA Diamond comparison in flight.
MTP-GGUF:
https://huggingface.co/ManniX-ITA/Qwen3.6-27B-Omnimerge-v4-MTP-GGUFOrganizations
None yet
ManniX-ITA/Qwen3.5-27B-Omnimerge-GGUF
27B β’ Updated β’ 190
ManniX-ITA/Qwen3.5-27B-Omnimerge
28B β’ Updated β’ 9
ManniX-ITA/gemma-4-A4B-109e-v3-it
22B β’ Updated β’ 6 β’ 3
ManniX-ITA/gemma-4-A4B-109e-it
22B β’ Updated β’ 1 β’ 1
ManniX-ITA/gemma-4-A4B-109e-it-GGUF
22B β’ Updated β’ 288 β’ 4
ManniX-ITA/llama3.1-8b-stormh3
Text Generation β’ 8B β’ Updated β’ 3
ManniX-ITA/llama3.1-8b-h3storm
Text Generation β’ 8B β’ Updated β’ 8
ManniX-ITA/Llama-3-8b-64k-PoSE-GGUF
Updated β’ 43
ManniX-ITA/dolphin-2.9-llama3-8b-dus-GGUF
10B β’ Updated β’ 64
ManniX-ITA/dolphin-2.9-llama3-8b-dus
Text Generation β’ 10B β’ Updated β’ 2
ManniX-ITA/testMoE
Text Generation β’ 24B β’ Updated β’ 6
ManniX-ITA/Starling-LM-7B-beta-LaserRMT-v1-GGUF
Updated β’ 23 β’ 2
ManniX-ITA/Starling-LM-7B-beta-LaserRMT-v1
Text Generation β’ 7B β’ Updated β’ 7 β’ 2
ManniX-ITA/Mixtral_7Bx2_MoE-GGUF
Text Generation β’ 13B β’ Updated β’ 214 β’ 3