r=128 Teacher-ASVD + Full Norm-Folding + CodeReadout)Rank-128 PEFT LoRA adapter and 255-way CodeReadout head extracted from fastino/GLiDE-27B onto Qwen/Qwen3.5-27B.
GLiDE-27B (passes=2) JevBench Accuracy (213 / 231 = 92.21%): Matches the full 54 GB fastino/GLiDE-27B checkpoint on JevBench (213/231, NLL = 0.1988 vs 0.1972, Brier = 0.1053 vs 0.1033) in a 1.6 GB LoRA adapter (+5 cases and -0.0180 Brier over unweighted Joint-SVD). Even in single-pass mode (passes=1, without midloop), it reaches 211 / 231 (91.34%, Brier = 0.1141).warp-34 + PEFT): Serve standard causal chat completions (lora_id = -1) and non-causal /v1/systemone structured decision queries (lora_id = "glide") from a single loaded Qwen3.5-27B backbone.lora_A Deduplication): Fused sibling projections (q_proj/k_proj/v_proj, gate_proj/up_proj, in_proj_qkv/in_proj_z, in_proj_b/in_proj_a) share an identical lora_A.weight computed via Joint Teacher-ASVD (128 calibration prompts from SargeDev/jev-distill-corpus-v3 with GLiDE-27B teacher activations), reducing fused LoRA A projection FLOPs and VRAM by 2x-3x when deduplicated by the inference engine.GemmaRMSNorm + RMSNormGated Folding: Pre-norm scale shifts (1 + g_glide) / (1 + g_base) (input_layernorm, post_attention_layernorm), linear_attn.norm (g_glide / g_base repeated across 48 value heads into linear_attn.out_proj), and final model.language_model.norm (folded into readout.safetensors) are absorbed into the adapter and readout weights with zero runtime overhead.231 Cases: 48 Easy / 72 Original / 111 Hard)Evaluated on kshitijthakkar/nirnaya-jevbench-public-eval-v1 using the native GLiDE-27B glide-v1 prompt template and noncausal_full_attention:
| Model | Overall Acc | Easy (48) |
Original (72) |
Hard (111) |
NLL ($\downarrow$) | Brier ($\downarrow$) |
|---|---|---|---|---|---|---|
Base Qwen/Qwen3.5-27B (passes=1 single-pass) |
204 / 231 (88.31%) |
48 / 48 (100.0%) |
69 / 72 (95.83%) |
87 / 111 (78.38%) |
0.3236 |
0.1738 |
Base Qwen/Qwen3.5-27B (passes=2 midloop) |
204 / 231 (88.31%) |
48 / 48 (100.0%) |
70 / 72 (97.22%) |
86 / 111 (77.48%) |
0.2947 |
0.1525 |
Qwen3.5-27B + GLiDE-LoRA (r=128 Joint-SVD, passes=1) |
206 / 231 (89.18%) |
48 / 48 (100.0%) |
70 / 72 (97.22%) |
88 / 111 (79.28%) |
0.2754 |
0.1522 |
Qwen3.5-27B + GLiDE-LoRA (r=128 Joint-SVD, passes=2) |
208 / 231 (90.04%) |
48 / 48 (100.0%) |
71 / 72 (98.61%) |
89 / 111 (80.18%) |
0.2275 |
0.1233 |
Qwen3.5-27B + GLiDE-LoRA (r=128 Teacher-ASVD, passes=1) |
211 / 231 (91.34%) |
48 / 48 (100.0%) |
71 / 72 (98.61%) |
92 / 111 (82.88%) |
0.2147 |
0.1141 |
Qwen3.5-27B + GLiDE-LoRA (r=128 Teacher-ASVD, passes=2) |
213 / 231 (92.21%) |
48 / 48 (100.0%) |
71 / 72 (98.61%) |
94 / 111 (84.68%) |
0.1988 |
0.1053 |
Full fastino/GLiDE-27B (passes=2 midloop) |
213 / 231 (92.21%) |
48 / 48 (100.0%) |
72 / 72 (100.0%) |
93 / 111 (83.78%) |
0.1972 |
0.1033 |
Full fastino/GLiDE-27B (passes=1 single-pass) |
214 / 231 (92.64%) |
48 / 48 (100.0%) |
72 / 72 (100.0%) |
94 / 111 (84.68%) |
0.1933 |
0.1022 |
adapter_config.json & adapter_model.safetensors: Standard PEFT LoRA weights (r=128, lora_alpha=128, rank_pattern & alpha_pattern = 96 on linear_attn.in_proj_{a,b}, bfloat16).decision_config.json: GLiDE-27B decision runtime metadata (attention_mode: "noncausal_full_attention", prompt_format: "glide-v1", midloop: {"blocks": [[24, 35], [36, 47]], "passes": 2}, 255 option codes A..GC).readout.safetensors: 255-way CodeReadout head ([255, 5120] in bfloat16, with final GemmaRMSNorm ratio folded in).adapter_model.safetensors 1.8 GBreadout.safetensors 3 MBQwen/Qwen3.5-27B