YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Qwen3.8-27B: matching Mia-AiLab EXL3 in GGUF (NVFPEXL + V1)

The experiment

The goal is to reproduce, in GGUF, the per-tensor size of Mia-AiLab's Qwen3.8-27B-EXL3-3.5bpw file. EXL3 is a single variable-bit format locked to the exllamav3/TabbyAPI runtime, so it cannot be loaded by llama.cpp. The method has two steps:

  1. NVFPEXL - map every EXL3 tensor to the closest ggml type for its declared bits: 4-bit -> NVFP4 (4.5 bpw), 3-bit -> Q3_K (3.4375), 5-bit -> Q5_K (5.5), 6-bit -> Q6_K (6.5625), 16-bit raw -> BF16 (16). This matches the shape of the EXL3 allocation tensor by tensor, but because ggml has no 3.0- or 4.0-bpw types it lands uniformly above EXL3 per class (the 4.5/3.4375 vs 4.0/3.0 gap).
  2. NVFPEXL-V1 - take the unsloth imatrix for the Qwen3.6-27B-MTP GGUF (496 projection weights ranked by activation energy/token over 849,561 calibration tokens) and demote low-importance tensors one step: the least-active NVFP4 -> Q3_K and the least-active Q3_K -> Q2_K (2.5625). This pushes the GGUF down toward, and partly below, the EXL3 size using the imatrix as an importance signal that EXL3 itself has no access to. The most active 70% of NVFP4 and 90% of Q3_K tensors by energy, all norms/aux, the MTP block (blk.64) and the LM head are left untouched.
File What it is Size
Qwen3.8-27B-EXL3-3.5bpw.mapping per-tensor map of the EXL3 snapshot (1199 tensors, HF names) 15,338,106,948 B (14.28 GiB) tensor data
Qwen3.8-27B-NVFPEXL.gguf closest-ggml-type build (step 1) 15,959,608,064 B (14.86 GiB)
Qwen3.8-27B-NVFPEXL-V1-imatrix.gguf imatrix-demoted build (step 2) 15,302,281,984 B (14.25 GiB)

Step 2 changed 96 of 866 tensors (88 NVFP4->Q3_K, 8 Q3_K->Q2_K), cutting 0.61 GiB (-4.12%) versus NVFPEXL.

Size summary

Grouped by the EXL3 declared-bit class, with the effective bpw and stored size of each of the three files. The text body is the same 866 tensors / 27,320,697,856 params in all three; the EXL3 file additionally carries a 333-tensor visual tower (0.86 GiB) that has no place in the text-only qwen35 GGUF.

EXL3 class tensors params EXL3 bpw EXL3 NVFPEXL bpw NVFPEXL V1 bpw V1
~4-bit 270 12,509,511,680 4.0049 5,972.30 MiB 4.5000 6,710.62 MiB 4.1259 6,152.81 MiB
~3-bit 137 12,210,667,520 3.0040 4,372.76 MiB 3.4375 5,003.71 MiB 3.3901 4,934.65 MiB
16-bit (raw) 457 1,297,662,976 16.0000 2,475.10 MiB 16.0329 2,480.19 MiB 16.0329 2,480.19 MiB
~6-bit 1 1,271,398,400 6.0032 909.86 MiB 6.5625 994.63 MiB 6.5625 994.63 MiB
~5-bit 1 31,457,280 5.0057 18.77 MiB 5.5000 20.62 MiB 5.5000 20.62 MiB
total (text body) 866 27,320,697,856 4.2215 13,748.79 MiB 4.6700 15,209.78 MiB 4.4776 14,582.91 MiB
whole file 27,781,427,952 (EXL3, incl. visual) / 27,320,697,856 (GGUF, text) 4.4168 14,627.56 MiB 4.6733 15,220.27 MiB 4.4808 14,593.39 MiB

Note on the 16-bit class: EXL3 stores all of it as BF16, but in the GGUF build 97 of the 457 tensors land as BF16 and 360 (norms and small aux vectors) as F32, matching how the source BF16 GGUF already stored them; that is why the GGUF side of the class is 16.033 bpw instead of 16.0.

At the file level the imatrix demotion brings V1 (4.4808 bpw) within 0.03 GiB of the whole EXL3 file (4.4168 bpw), while NVFPEXL without demotion sits +10.6% over the EXL3 text body. Note the EXL3 whole-file bpw includes the visual tower, so the fair like-for-like comparison is the text-body row: NVFPEXL +10.6%, V1 +6.1% versus the EXL3 text body.

Per-tensor comparison

All 866 tensors in GGUF file order. The EXL3 column is the declared codebook depth (4b/3b/5b/6b = variable-bit quant, 16b = stored raw); NVFPEXL and V1 show the ggml type in each built file. dV1-V0 = V1 minus NVFPEXL (the imatrix demotion, MiB); dV1-EXL3 = V1 minus EXL3 (distance from the target, MiB; negative means V1 is smaller than EXL3 for that tensor).

# tensor params EXL3 NVFPEXL V1 dV1-V0 (MiB) dV1-EXL3 (MiB)
1 output.weight 1,271,398,400 6b Q6_K Q6_K +0.00 +84.77
2 token_embd.weight 1,271,398,400 16b BF16 BF16 +0.00 +0.00
3 blk.0.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
4 blk.0.ssm_a 48 16b F32 F32 +0.00 +0.00
5 blk.0.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
6 blk.0.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
7 blk.0.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
8 blk.0.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
9 blk.0.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
10 blk.0.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
11 blk.0.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
12 blk.0.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
13 blk.0.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
14 blk.0.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
15 blk.0.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
16 blk.0.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
17 blk.1.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
18 blk.1.ssm_a 48 16b F32 F32 +0.00 +0.00
19 blk.1.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
20 blk.1.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
21 blk.1.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
22 blk.1.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
23 blk.1.attn_qkv.weight 52,428,800 4b NVFP4 Q3_K -6.64 -3.54
24 blk.1.attn_gate.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
25 blk.1.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
26 blk.1.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
27 blk.1.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
28 blk.1.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
29 blk.1.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
30 blk.1.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
31 blk.10.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
32 blk.10.ssm_a 48 16b F32 F32 +0.00 +0.00
33 blk.10.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
34 blk.10.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
35 blk.10.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
36 blk.10.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
37 blk.10.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
38 blk.10.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
39 blk.10.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
40 blk.10.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
41 blk.10.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
42 blk.10.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
43 blk.10.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
44 blk.10.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
45 blk.11.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
46 blk.11.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
47 blk.11.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
48 blk.11.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
49 blk.11.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
50 blk.11.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
51 blk.11.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
52 blk.11.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
53 blk.11.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
54 blk.11.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
55 blk.11.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
56 blk.12.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
57 blk.12.ssm_a 48 16b F32 F32 +0.00 +0.00
58 blk.12.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
59 blk.12.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
60 blk.12.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
61 blk.12.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
62 blk.12.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
63 blk.12.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
64 blk.12.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
65 blk.12.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
66 blk.12.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
67 blk.12.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
68 blk.12.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
69 blk.12.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
70 blk.13.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
71 blk.13.ssm_a 48 16b F32 F32 +0.00 +0.00
72 blk.13.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
73 blk.13.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
74 blk.13.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
75 blk.13.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
76 blk.13.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
77 blk.13.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
78 blk.13.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
79 blk.13.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
80 blk.13.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
81 blk.13.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
82 blk.13.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
83 blk.13.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
84 blk.14.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
85 blk.14.ssm_a 48 16b F32 F32 +0.00 +0.00
86 blk.14.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
87 blk.14.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
88 blk.14.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
89 blk.14.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
90 blk.14.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
91 blk.14.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
92 blk.14.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
93 blk.14.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
94 blk.14.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
95 blk.14.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
96 blk.14.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
97 blk.14.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
98 blk.15.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
99 blk.15.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
100 blk.15.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
101 blk.15.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
102 blk.15.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
103 blk.15.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
104 blk.15.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
105 blk.15.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
106 blk.15.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
107 blk.15.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
108 blk.15.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
109 blk.16.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
110 blk.16.ssm_a 48 16b F32 F32 +0.00 +0.00
111 blk.16.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
112 blk.16.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
113 blk.16.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
114 blk.16.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
115 blk.16.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
116 blk.16.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
117 blk.16.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
118 blk.16.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
119 blk.16.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
120 blk.16.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
121 blk.16.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
122 blk.16.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
123 blk.17.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
124 blk.17.ssm_a 48 16b F32 F32 +0.00 +0.00
125 blk.17.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
126 blk.17.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
127 blk.17.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
128 blk.17.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
129 blk.17.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
130 blk.17.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
131 blk.17.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
132 blk.17.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
133 blk.17.ffn_down.weight 89,128,960 3b Q3_K Q2_K -8.63 -4.03
134 blk.17.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
135 blk.17.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
136 blk.17.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
137 blk.18.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
138 blk.18.ssm_a 48 16b F32 F32 +0.00 +0.00
139 blk.18.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
140 blk.18.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
141 blk.18.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
142 blk.18.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
143 blk.18.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
144 blk.18.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
145 blk.18.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
146 blk.18.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
147 blk.18.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
148 blk.18.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
149 blk.18.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
150 blk.18.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
151 blk.19.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
152 blk.19.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
153 blk.19.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
154 blk.19.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
155 blk.19.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
156 blk.19.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
157 blk.19.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
158 blk.19.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
159 blk.19.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
160 blk.19.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
161 blk.19.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
162 blk.2.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
163 blk.2.ssm_a 48 16b F32 F32 +0.00 +0.00
164 blk.2.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
165 blk.2.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
166 blk.2.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
167 blk.2.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
168 blk.2.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
169 blk.2.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
170 blk.2.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
171 blk.2.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
172 blk.2.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
173 blk.2.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
174 blk.2.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
175 blk.2.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
176 blk.20.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
177 blk.20.ssm_a 48 16b F32 F32 +0.00 +0.00
178 blk.20.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
179 blk.20.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
180 blk.20.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
181 blk.20.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
182 blk.20.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
183 blk.20.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
184 blk.20.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
185 blk.20.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
186 blk.20.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
187 blk.20.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
188 blk.20.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
189 blk.20.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
190 blk.21.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
191 blk.21.ssm_a 48 16b F32 F32 +0.00 +0.00
192 blk.21.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
193 blk.21.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
194 blk.21.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
195 blk.21.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
196 blk.21.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
197 blk.21.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
198 blk.21.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
199 blk.21.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
200 blk.21.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
201 blk.21.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
202 blk.21.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
203 blk.21.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
204 blk.22.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
205 blk.22.ssm_a 48 16b F32 F32 +0.00 +0.00
206 blk.22.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
207 blk.22.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
208 blk.22.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
209 blk.22.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
210 blk.22.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
211 blk.22.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
212 blk.22.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
213 blk.22.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
214 blk.22.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
215 blk.22.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
216 blk.22.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
217 blk.22.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
218 blk.23.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
219 blk.23.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
220 blk.23.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
221 blk.23.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
222 blk.23.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
223 blk.23.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
224 blk.23.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
225 blk.23.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
226 blk.23.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
227 blk.23.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
228 blk.23.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
229 blk.24.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
230 blk.24.ssm_a 48 16b F32 F32 +0.00 +0.00
231 blk.24.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
232 blk.24.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
233 blk.24.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
234 blk.24.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
235 blk.24.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
236 blk.24.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
237 blk.24.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
238 blk.24.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
239 blk.24.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
240 blk.24.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
241 blk.24.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
242 blk.24.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
243 blk.25.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
244 blk.25.ssm_a 48 16b F32 F32 +0.00 +0.00
245 blk.25.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
246 blk.25.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
247 blk.25.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
248 blk.25.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
249 blk.25.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
250 blk.25.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
251 blk.25.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
252 blk.25.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
253 blk.25.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
254 blk.25.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
255 blk.25.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
256 blk.25.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
257 blk.26.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
258 blk.26.ssm_a 48 16b F32 F32 +0.00 +0.00
259 blk.26.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
260 blk.26.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
261 blk.26.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
262 blk.26.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
263 blk.26.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
264 blk.26.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
265 blk.26.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
266 blk.26.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
267 blk.26.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
268 blk.26.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
269 blk.26.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
270 blk.26.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
271 blk.27.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
272 blk.27.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
273 blk.27.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
274 blk.27.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
275 blk.27.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
276 blk.27.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
277 blk.27.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
278 blk.27.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
279 blk.27.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
280 blk.27.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
281 blk.27.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
282 blk.28.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
283 blk.28.ssm_a 48 16b F32 F32 +0.00 +0.00
284 blk.28.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
285 blk.28.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
286 blk.28.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
287 blk.28.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
288 blk.28.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
289 blk.28.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
290 blk.28.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
291 blk.28.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
292 blk.28.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
293 blk.28.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
294 blk.28.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
295 blk.28.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
296 blk.29.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
297 blk.29.ssm_a 48 16b F32 F32 +0.00 +0.00
298 blk.29.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
299 blk.29.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
300 blk.29.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
301 blk.29.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
302 blk.29.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
303 blk.29.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
304 blk.29.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
305 blk.29.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
306 blk.29.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
307 blk.29.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
308 blk.29.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
309 blk.29.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
310 blk.3.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
311 blk.3.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
312 blk.3.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
313 blk.3.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
314 blk.3.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
315 blk.3.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
316 blk.3.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
317 blk.3.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
318 blk.3.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
319 blk.3.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
320 blk.3.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
321 blk.30.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
322 blk.30.ssm_a 48 16b F32 F32 +0.00 +0.00
323 blk.30.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
324 blk.30.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
325 blk.30.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
326 blk.30.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
327 blk.30.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
328 blk.30.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
329 blk.30.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
330 blk.30.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
331 blk.30.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
332 blk.30.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
333 blk.30.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
334 blk.30.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
335 blk.31.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
336 blk.31.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
337 blk.31.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
338 blk.31.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
339 blk.31.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
340 blk.31.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
341 blk.31.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
342 blk.31.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
343 blk.31.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
344 blk.31.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
345 blk.31.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
346 blk.32.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
347 blk.32.ssm_a 48 16b F32 F32 +0.00 +0.00
348 blk.32.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
349 blk.32.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
350 blk.32.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
351 blk.32.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
352 blk.32.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
353 blk.32.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
354 blk.32.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
355 blk.32.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
356 blk.32.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
357 blk.32.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
358 blk.32.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
359 blk.32.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
360 blk.33.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
361 blk.33.ssm_a 48 16b F32 F32 +0.00 +0.00
362 blk.33.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
363 blk.33.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
364 blk.33.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
365 blk.33.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
366 blk.33.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
367 blk.33.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
368 blk.33.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
369 blk.33.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
370 blk.33.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
371 blk.33.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
372 blk.33.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
373 blk.33.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
374 blk.34.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
375 blk.34.ssm_a 48 16b F32 F32 +0.00 +0.00
376 blk.34.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
377 blk.34.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
378 blk.34.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
379 blk.34.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
380 blk.34.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
381 blk.34.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
382 blk.34.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
383 blk.34.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
384 blk.34.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
385 blk.34.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
386 blk.34.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
387 blk.34.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
388 blk.35.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
389 blk.35.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
390 blk.35.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
391 blk.35.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
392 blk.35.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
393 blk.35.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
394 blk.35.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
395 blk.35.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
396 blk.35.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
397 blk.35.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
398 blk.35.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
399 blk.36.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
400 blk.36.ssm_a 48 16b F32 F32 +0.00 +0.00
401 blk.36.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
402 blk.36.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
403 blk.36.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
404 blk.36.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
405 blk.36.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
406 blk.36.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
407 blk.36.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
408 blk.36.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
409 blk.36.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
410 blk.36.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
411 blk.36.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
412 blk.36.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
413 blk.37.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
414 blk.37.ssm_a 48 16b F32 F32 +0.00 +0.00
415 blk.37.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
416 blk.37.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
417 blk.37.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
418 blk.37.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
419 blk.37.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
420 blk.37.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
421 blk.37.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
422 blk.37.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
423 blk.37.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
424 blk.37.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
425 blk.37.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
426 blk.37.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
427 blk.38.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
428 blk.38.ssm_a 48 16b F32 F32 +0.00 +0.00
429 blk.38.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
430 blk.38.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
431 blk.38.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
432 blk.38.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
433 blk.38.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
434 blk.38.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
435 blk.38.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
436 blk.38.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
437 blk.38.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
438 blk.38.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
439 blk.38.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
440 blk.38.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
441 blk.39.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
442 blk.39.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
443 blk.39.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
444 blk.39.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
445 blk.39.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
446 blk.39.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
447 blk.39.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
448 blk.39.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
449 blk.39.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
450 blk.39.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
451 blk.39.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
452 blk.4.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
453 blk.4.ssm_a 48 16b F32 F32 +0.00 +0.00
454 blk.4.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
455 blk.4.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
456 blk.4.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
457 blk.4.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
458 blk.4.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
459 blk.4.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
460 blk.4.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
461 blk.4.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
462 blk.4.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
463 blk.4.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
464 blk.4.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
465 blk.4.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
466 blk.40.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
467 blk.40.ssm_a 48 16b F32 F32 +0.00 +0.00
468 blk.40.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
469 blk.40.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
470 blk.40.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
471 blk.40.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
472 blk.40.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
473 blk.40.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
474 blk.40.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
475 blk.40.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
476 blk.40.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
477 blk.40.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
478 blk.40.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
479 blk.40.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
480 blk.41.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
481 blk.41.ssm_a 48 16b F32 F32 +0.00 +0.00
482 blk.41.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
483 blk.41.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
484 blk.41.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
485 blk.41.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
486 blk.41.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
487 blk.41.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
488 blk.41.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
489 blk.41.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
490 blk.41.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
491 blk.41.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
492 blk.41.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
493 blk.41.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
494 blk.42.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
495 blk.42.ssm_a 48 16b F32 F32 +0.00 +0.00
496 blk.42.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
497 blk.42.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
498 blk.42.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
499 blk.42.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
500 blk.42.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
501 blk.42.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
502 blk.42.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
503 blk.42.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
504 blk.42.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
505 blk.42.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
506 blk.42.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
507 blk.42.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
508 blk.43.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
509 blk.43.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
510 blk.43.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
511 blk.43.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
512 blk.43.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
513 blk.43.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
514 blk.43.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
515 blk.43.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
516 blk.43.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
517 blk.43.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
518 blk.43.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
519 blk.44.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
520 blk.44.ssm_a 48 16b F32 F32 +0.00 +0.00
521 blk.44.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
522 blk.44.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
523 blk.44.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
524 blk.44.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
525 blk.44.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
526 blk.44.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
527 blk.44.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
528 blk.44.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
529 blk.44.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
530 blk.44.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
531 blk.44.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
532 blk.44.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
533 blk.45.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
534 blk.45.ssm_a 48 16b F32 F32 +0.00 +0.00
535 blk.45.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
536 blk.45.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
537 blk.45.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
538 blk.45.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
539 blk.45.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
540 blk.45.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
541 blk.45.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
542 blk.45.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
543 blk.45.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
544 blk.45.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
545 blk.45.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
546 blk.45.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
547 blk.46.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
548 blk.46.ssm_a 48 16b F32 F32 +0.00 +0.00
549 blk.46.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
550 blk.46.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
551 blk.46.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
552 blk.46.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
553 blk.46.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
554 blk.46.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
555 blk.46.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
556 blk.46.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
557 blk.46.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
558 blk.46.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
559 blk.46.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
560 blk.46.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
561 blk.47.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
562 blk.47.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
563 blk.47.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
564 blk.47.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
565 blk.47.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
566 blk.47.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
567 blk.47.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
568 blk.47.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
569 blk.47.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
570 blk.47.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
571 blk.47.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
572 blk.48.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
573 blk.48.ssm_a 48 16b F32 F32 +0.00 +0.00
574 blk.48.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
575 blk.48.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
576 blk.48.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
577 blk.48.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
578 blk.48.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
579 blk.48.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
580 blk.48.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
581 blk.48.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
582 blk.48.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
583 blk.48.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
584 blk.48.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
585 blk.48.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
586 blk.49.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
587 blk.49.ssm_a 48 16b F32 F32 +0.00 +0.00
588 blk.49.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
589 blk.49.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
590 blk.49.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
591 blk.49.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
592 blk.49.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
593 blk.49.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
594 blk.49.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
595 blk.49.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
596 blk.49.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
597 blk.49.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
598 blk.49.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
599 blk.49.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
600 blk.5.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
601 blk.5.ssm_a 48 16b F32 F32 +0.00 +0.00
602 blk.5.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
603 blk.5.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
604 blk.5.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
605 blk.5.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
606 blk.5.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
607 blk.5.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
608 blk.5.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
609 blk.5.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
610 blk.5.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
611 blk.5.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
612 blk.5.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
613 blk.5.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
614 blk.50.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
615 blk.50.ssm_a 48 16b F32 F32 +0.00 +0.00
616 blk.50.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
617 blk.50.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
618 blk.50.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
619 blk.50.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
620 blk.50.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
621 blk.50.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
622 blk.50.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
623 blk.50.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
624 blk.50.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
625 blk.50.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
626 blk.50.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
627 blk.50.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
628 blk.51.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
629 blk.51.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
630 blk.51.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
631 blk.51.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
632 blk.51.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
633 blk.51.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
634 blk.51.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
635 blk.51.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
636 blk.51.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
637 blk.51.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
638 blk.51.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
639 blk.52.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
640 blk.52.ssm_a 48 16b F32 F32 +0.00 +0.00
641 blk.52.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
642 blk.52.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
643 blk.52.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
644 blk.52.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
645 blk.52.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
646 blk.52.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
647 blk.52.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
648 blk.52.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
649 blk.52.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
650 blk.52.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
651 blk.52.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
652 blk.52.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
653 blk.53.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
654 blk.53.ssm_a 48 16b F32 F32 +0.00 +0.00
655 blk.53.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
656 blk.53.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
657 blk.53.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
658 blk.53.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
659 blk.53.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
660 blk.53.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
661 blk.53.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
662 blk.53.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
663 blk.53.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
664 blk.53.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
665 blk.53.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
666 blk.53.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
667 blk.54.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
668 blk.54.ssm_a 48 16b F32 F32 +0.00 +0.00
669 blk.54.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
670 blk.54.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
671 blk.54.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
672 blk.54.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
673 blk.54.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
674 blk.54.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
675 blk.54.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
676 blk.54.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
677 blk.54.ffn_down.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
678 blk.54.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
679 blk.54.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
680 blk.54.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
681 blk.55.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
682 blk.55.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
683 blk.55.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
684 blk.55.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
685 blk.55.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
686 blk.55.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
687 blk.55.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
688 blk.55.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
689 blk.55.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
690 blk.55.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
691 blk.55.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
692 blk.56.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
693 blk.56.ssm_a 48 16b F32 F32 +0.00 +0.00
694 blk.56.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
695 blk.56.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
696 blk.56.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
697 blk.56.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
698 blk.56.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
699 blk.56.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
700 blk.56.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
701 blk.56.ssm_out.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
702 blk.56.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
703 blk.56.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
704 blk.56.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
705 blk.56.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
706 blk.57.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
707 blk.57.ssm_a 48 16b F32 F32 +0.00 +0.00
708 blk.57.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
709 blk.57.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
710 blk.57.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
711 blk.57.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
712 blk.57.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
713 blk.57.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
714 blk.57.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
715 blk.57.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
716 blk.57.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
717 blk.57.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
718 blk.57.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
719 blk.57.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
720 blk.58.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
721 blk.58.ssm_a 48 16b F32 F32 +0.00 +0.00
722 blk.58.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
723 blk.58.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
724 blk.58.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
725 blk.58.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
726 blk.58.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
727 blk.58.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
728 blk.58.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
729 blk.58.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
730 blk.58.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
731 blk.58.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
732 blk.58.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
733 blk.58.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
734 blk.59.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
735 blk.59.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
736 blk.59.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
737 blk.59.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
738 blk.59.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
739 blk.59.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
740 blk.59.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
741 blk.59.attn_output.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
742 blk.59.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
743 blk.59.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
744 blk.59.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
745 blk.6.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
746 blk.6.ssm_a 48 16b F32 F32 +0.00 +0.00
747 blk.6.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
748 blk.6.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
749 blk.6.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
750 blk.6.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
751 blk.6.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
752 blk.6.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
753 blk.6.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
754 blk.6.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
755 blk.6.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
756 blk.6.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
757 blk.6.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
758 blk.6.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
759 blk.60.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
760 blk.60.ssm_a 48 16b F32 F32 +0.00 +0.00
761 blk.60.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
762 blk.60.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
763 blk.60.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
764 blk.60.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
765 blk.60.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
766 blk.60.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
767 blk.60.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
768 blk.60.ssm_out.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
769 blk.60.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
770 blk.60.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
771 blk.60.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
772 blk.60.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
773 blk.61.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
774 blk.61.ssm_a 48 16b F32 F32 +0.00 +0.00
775 blk.61.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
776 blk.61.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
777 blk.61.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
778 blk.61.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
779 blk.61.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
780 blk.61.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
781 blk.61.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
782 blk.61.ssm_out.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
783 blk.61.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
784 blk.61.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
785 blk.61.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
786 blk.61.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
787 blk.62.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
788 blk.62.ssm_a 48 16b F32 F32 +0.00 +0.00
789 blk.62.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
790 blk.62.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
791 blk.62.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
792 blk.62.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
793 blk.62.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
794 blk.62.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
795 blk.62.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
796 blk.62.ssm_out.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
797 blk.62.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
798 blk.62.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
799 blk.62.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
800 blk.62.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
801 blk.63.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
802 blk.63.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
803 blk.63.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
804 blk.63.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
805 blk.63.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
806 blk.63.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
807 blk.63.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
808 blk.63.attn_output.weight 31,457,280 5b Q5_K Q5_K +0.00 +1.85
809 blk.63.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
810 blk.63.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
811 blk.63.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
812 blk.7.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
813 blk.7.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
814 blk.7.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
815 blk.7.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
816 blk.7.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
817 blk.7.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
818 blk.7.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
819 blk.7.attn_output.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
820 blk.7.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
821 blk.7.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
822 blk.7.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
823 blk.8.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
824 blk.8.ssm_a 48 16b F32 F32 +0.00 +0.00
825 blk.8.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
826 blk.8.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
827 blk.8.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
828 blk.8.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
829 blk.8.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
830 blk.8.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
831 blk.8.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
832 blk.8.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
833 blk.8.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
834 blk.8.ffn_gate.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
835 blk.8.ffn_up.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
836 blk.8.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
837 blk.9.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
838 blk.9.ssm_a 48 16b F32 F32 +0.00 +0.00
839 blk.9.ssm_conv1d.weight 40,960 16b F32 F32 +0.00 +0.08
840 blk.9.ssm_dt.bias 48 16b F32 F32 +0.00 +0.00
841 blk.9.ssm_alpha.weight 245,760 16b BF16 BF16 +0.00 +0.00
842 blk.9.ssm_beta.weight 245,760 16b BF16 BF16 +0.00 +0.00
843 blk.9.attn_qkv.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
844 blk.9.attn_gate.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
845 blk.9.ssm_norm.weight 128 16b F32 F32 +0.00 +0.00
846 blk.9.ssm_out.weight 31,457,280 4b NVFP4 Q3_K -3.98 -2.13
847 blk.9.ffn_down.weight 89,128,960 4b NVFP4 Q3_K -11.29 -6.02
848 blk.9.ffn_gate.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
849 blk.9.ffn_up.weight 89,128,960 3b Q3_K Q3_K +0.00 +4.61
850 blk.9.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
851 output_norm.weight 5,120 16b F32 F32 +0.00 +0.01
852 blk.64.nextn.eh_proj.weight 52,428,800 4b NVFP4 NVFP4 +0.00 +3.10
853 blk.64.attn_norm.weight 5,120 16b F32 F32 +0.00 +0.01
854 blk.64.ffn_down.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
855 blk.64.ffn_gate.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
856 blk.64.ffn_up.weight 89,128,960 4b NVFP4 NVFP4 +0.00 +5.27
857 blk.64.post_attention_norm.weight 5,120 16b F32 F32 +0.00 +0.01
858 blk.64.attn_k_norm.weight 256 16b F32 F32 +0.00 +0.00
859 blk.64.attn_k.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
860 blk.64.attn_output.weight 31,457,280 4b NVFP4 NVFP4 +0.00 +1.85
861 blk.64.attn_q_norm.weight 256 16b F32 F32 +0.00 +0.00
862 blk.64.attn_q.weight 62,914,560 4b NVFP4 NVFP4 +0.00 +3.72
863 blk.64.attn_v.weight 5,242,880 4b NVFP4 NVFP4 +0.00 +0.30
864 blk.64.nextn.shared_head_norm.weight 5,120 16b F32 F32 +0.00 +0.01
865 blk.64.nextn.enorm.weight 5,120 16b F32 F32 +0.00 +0.01
866 blk.64.nextn.hnorm.weight 5,120 16b F32 F32 +0.00 +0.01

Visual tower dropped from the GGUF: 333 tensors, 460,730,096 params, 921,460,192 B (0.86 GiB). The qwen35 arch is text-only; vision runs through a separate mmproj.

Build

cd /src/quant && uv run quant quantize \
  --mapping  exl3/quant/Qwen3.8-27B-NVFPEXL-V1-imatrix-gguf.mapping \
  --source   /src/quant/artifacts/Qwen3.8-27B-BF16.gguf \
  --output   exl3/quant/Qwen3.8-27B-NVFPEXL-V1-imatrix.gguf \
  --placeholder Q8_0 -t 32
Downloads last month
337
GGUF
Model size
27B params
Architecture
qwen35
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support