Axion-1.0 35B - Q8_0

APEX Quantization Series by IsValorum

[!IMPORTANT] Independently computed using dedicated cloud compute clusters, funded 100% out-of-pocket.

If you consider these releases to be of high quality and they help you in your day-to-day workflow, consider supporting on Ko-fi. Every contribution directly sustains future cloud compute hours and helps support my university studies.

Family cross-links

Tier	Footprint	Base	Abliterated
Q8_0	36.9 GB	Axion-1.0-35B	—

Architecture and precision

Base model: Axion1-0/Axion-1.0-35B
Architecture: MoE — 40 layers, 30 linear-attention/SSM + 10 full-attention; 256 routed experts + 1 shared, 8 active; hidden 2048; 262,144-token context; 248,320 vocab
iMatrix: not used — plain Q8_0; this file is the calibration source for the family I-Matrix
Recurrent shielding: 1D convs, state params, routers and RMSNorms in F32

Empirical evaluation metrics

Variant	Perplexity	Delta vs. BF16	Fidelity retention
Base BF16	pending	pending	pending
Q8_0	pending	pending	pending

Audited tensor map

Tensor group	Layers	Count	Precision
`blk.*.attn_gate.weight`	SSM layers except 0	29	Q8_0
`blk.*.attn_k.weight`	full attention 3,7,...,39	10	Q8_0
`blk.*.attn_k_norm.weight`	full attention 3,7,...,39	10	F32
`blk..attn_norm.weight, blk..post_attention_norm.weight`	all 40	80	F32
`blk.*.attn_output.weight`	full attention 3,7,...,39	10	Q8_0
`blk.*.attn_q.weight`	full attention 3,7,...,39	10	Q8_0
`blk.*.attn_q_norm.weight`	full attention 3,7,...,39	10	F32
`blk.*.attn_qkv.weight`	SSM layers except 0	29	Q8_0
`blk.*.attn_v.weight`	full attention 3,7,...,39	10	Q8_0
`blk.*.ffn_down_exps.weight`	0-9	10	Q8_0
`blk.*.ffn_down_exps.weight`	10-29	20	Q8_0
`blk.*.ffn_down_exps.weight`	30-39	10	Q8_0
`blk.*.ffn_down_shexp.weight`	all 40	40	Q8_0
`blk.*.ffn_gate_exps.weight`	0-9	10	Q8_0
`blk.*.ffn_gate_exps.weight`	10-29	20	Q8_0
`blk.*.ffn_gate_exps.weight`	30-39	10	Q8_0
`blk.*.ffn_gate_inp.weight`	all 40	40	F32
`blk.*.ffn_gate_inp_shexp.weight`	all 40	40	F32
`blk.*.ffn_gate_shexp.weight`	all 40	40	Q8_0
`blk.*.ffn_up_exps.weight`	0-9	10	Q8_0
`blk.*.ffn_up_exps.weight`	10-29	20	Q8_0
`blk.*.ffn_up_exps.weight`	30-39	10	Q8_0
`blk.*.ffn_up_shexp.weight`	all 40	40	Q8_0
`blk.*.ssm_a`	SSM layers	30	F32
`blk.*.ssm_alpha.weight`	SSM layers	30	Q8_0
`blk.*.ssm_beta.weight`	SSM layers except 0	29	Q8_0
`blk.*.ssm_conv1d.weight`	SSM layers	30	F32
`blk.*.ssm_dt.bias`	SSM layers	30	F32
`blk.*.ssm_norm.weight`	SSM layers	30	F32
`blk.*.ssm_out.weight`	SSM layers except 0	29	Q8_0
`blk.0.attn_gate.weight`	layer 0	1	Q8_0
`blk.0.attn_qkv.weight`	layer 0	1	Q8_0
`blk.0.ssm_beta.weight`	layer 0	1	Q8_0
`blk.0.ssm_out.weight`	layer 0	1	Q8_0
`output.weight`	global	1	Q8_0
`output_norm.weight`	global	1	F32
`token_embd.weight`	global	1	Q8_0
The complete audited list is `tensor-map.tsv`.

IsValorum/Axion-1.0-35B-Q8_0-GGUF

Axion-1.0 35B - Q8_0

Family cross-links

Architecture and precision

Empirical evaluation metrics

Audited tensor map