加载中…
加载中…
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.
| Tier | Footprint | Base | Abliterated |
|---|---|---|---|
| MiniPlus | 12.03 GiB | Base | — |
| NanoPlus | 9.92 GiB | Current | — |
bottlecapai/ThinkingCap-gemma-4-26B-A4B-itgemma4model:bottlecapai/ThinkingCap-gemma-4-26B-A4B-it-GGUF/ThinkingCap-gemma-4-26B-A4B-it-imatrix.gguf@bb52b322579d218a9931261d6c9cd7403d80eb96 (used file; external provenance not inferred)The table below is generated from the audited physical GGUF.
| Layer Group | Sub-Component / Tensor | Qty | Precision | Engineering Rationale |
|---|---|---|---|---|
| Attention and gates | blk.[N].attn_k.weight |
30 | Q4_K |
Audited attention projection |
| Attention and gates | blk.[N].attn_output.weight |
25 | Q4_K |
Audited attention projection |
| Attention and gates | blk.[N].attn_output.weight |
5 | Q6_K |
Audited attention projection |
| Attention and gates | blk.[N].attn_q.weight |
30 | Q4_K |
Audited attention projection |
| Attention and gates | blk.[N].attn_v.weight |
25 | Q4_K |
Audited attention projection |
| Dense FFN / MLP | blk.[N].ffn_down.weight |
30 | IQ4_NL |
Audited dense MLP projection |
| Dense FFN / MLP | blk.[N].ffn_gate.weight |
30 | Q4_K |
Audited dense MLP projection |
| Dense FFN / MLP | blk.[N].ffn_up.weight |
30 | Q4_K |
Audited dense MLP projection |
| Global tensors | rope_freqs.weight |
1 | F32 |
Audited global tensor |
| Global tensors | token_embd.weight |
1 | Q6_K |
Audited global tensor |
| Normalizations | blk.[N].attn_k_norm.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].attn_norm.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].attn_q_norm.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].ffn_norm.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].post_attention_norm.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].post_ffw_norm.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].post_ffw_norm_1.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].post_ffw_norm_2.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | blk.[N].pre_ffw_norm_2.weight |
30 | F32 |
Verified physical normalization type |
| Normalizations | output_norm.weight |
1 | F32 |
Verified physical normalization type |
| Other block components | blk.[N].layer_output_scale.weight |
30 | F32 |
Audited structural parameter |
| Routed experts | blk.[N].ffn_down_exps.scale |
30 | F32 |
Audited routed-expert projection |
| Routed experts | blk.[N].ffn_down_exps.weight |
26 | IQ4_NL |
Audited routed-expert projection |
| Routed experts | blk.[N].ffn_down_exps.weight |
4 | Q5_0 |
Audited routed-expert projection |
| Routed experts | blk.[N].ffn_gate_up_exps.weight |
14 | IQ2_S |
Audited routed-expert projection |
| Routed experts | blk.[N].ffn_gate_up_exps.weight |
12 | IQ2_XXS |
Audited routed-expert projection |
| Routed experts | blk.[N].ffn_gate_up_exps.weight |
4 | IQ3_XXS |
Audited routed-expert projection |
| Routing and indexer | blk.[N].ffn_gate_inp.scale |
30 | F32 |
Audited routing/indexer parameter |
| Routing and indexer | blk.[N].ffn_gate_inp.weight |
30 | F32 |
Audited routing/indexer parameter |