Quick Navigation Index
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.
- Optimization History & Transparency Notice
- Quality Spectrum: APEX-I-NanoPlus vs. Standard Flat Quantizations
- Model Files & Technical Specifications
- Surgical Tensor Quantization Map (Audited from GGUF)
- Inference Quickstart
- 1. llama-cli (Console Generation)
- 2. llama-server (OpenAI-Compatible API)
- CRITICAL: Coding Syntax & Repeat Penalty Advisory (Preventing Character Swapping)
- Hardened Agentic Chat Template & Reasoning Effort
- Optional Support
Occamy-1.0 APEX-I-NanoPlus-V2.1 Abliterated GGUF (Uncensored / Zero Refusal)
The Next-Generation Frontier MoE · Extreme 12–13GB Footprint · Fast System RAM Streaming & Massive Context on 16GB VRAM
[!NOTE]
EXPLORE THE ESTABLISHED 35B MoE MINIPLUS & NANOPLUS LINEUP
These are complementary APEX-I releases, not alternate downloads of the same model. Each receives the same surgical tensor-by-tensor approach and a design suitable for full or partial system-RAM inference:
- Occamy-1.0 APEX-I-NanoPlus — the standard aligned NanoPlus release for broad reasoning and multimodal workflows.
- Occamy-1.0 APEX-I-MiniPlus-V2.1 Abliterated — the companion V2.1 refusal-ablated edition (14.75 GB / 13.74 GiB) delivering Q5_K_M tier fidelity with a strict 0% 2-bit quantization policy for 24GB+ workstations.
- Occamy-1.0 APEX-I-MiniPlus-V2.1 — the standard safety-aligned edition with default guardrails.
[!IMPORTANT]
THE DEFINITIVE SPECIFICATION IN THE approx. 12–13 GB CEILING
This APEX-I-NanoPlus-V2.1 Abliterated release delivers an ultra-compact architectural configuration for sparse Mixture-of-Experts quantization within an extreme approx. 12–13 GB envelope (12.55 GB / 11.69 GiB, 2.90 BPW). Every tensor across its 40 layers and 256 micro-experts has been mathematically allocated to maximize reasoning precision, preserve routing behavior, and eliminate AVX2 CPU dequantization stalls during hybrid and system-RAM inference.
[!TIP]
AUTHENTIC DIRECTIONAL ABLITERATION (UNCENSORED)
This is the official Abliterated / Uncensored edition of Occamy-1.0 APEX-I-NanoPlus V2.1, quantized directly from the authentic unquantized directional source with importance matrix (
imatrix) calibration:
- Zero Moralizing Refusals: Complete elimination of refusal vectors across cybersecurity, penetration testing, compliance, and systems programming tasks.
- Orthogonal Activation Steering: Refusal directions are isolated orthogonally to preserve 100% of domain knowledge, mathematical deduction, and MoE routing logic.
- 16GB VRAM Optimization: Leaves over 3 GB of free VRAM on 16GB consumer graphics cards (such as RTX 4080 and RTX 4070 Ti Super) for deep 32K context windows without GPU out-of-memory errors.
[!TIP]
BUILD & VERIFIED REFERENCE COMPARISON
Quantization Specification File Size (Disk) Memory Footprint (RAM/VRAM) Average BPW WikiText-2 Perplexity Quality Tier Equivalent Unquantized BF16 Base approx. 70.0 GB approx. 65.2 GiB 16.00 BPW approx. 6.18 (Reference) Full precision baseline APEX-I-MiniPlus V2.1 Abliterated 14.75 GB 13.74 GiB 3.52 BPW approx. 6.24 (Reference) Near-lossless Q5_K_M tier (0% 2-bit policy) APEX-I-NanoPlus-V2.1 Abliterated (CURRENT) 12.55 GB 11.69 GiB 2.90 BPW Calibrated near baseline Solid Q4_K_M Tier at 82.1% weight reduction Looking for the uncompromised 0% 2-bit tier? APEX-I-MiniPlus V2.1 Abliterated offers the 14.75 GB (3.52 BPW) release, protecting 100% of all expert weights above 3 bits.
Optimization History & Transparency Notice
We maintain our releases publicly as a transparent engineering record of continuous optimization:
| Specification | Core Experts (Layers 2–37) | Edge Experts (Layers 0–1, 38–39) | Shared Expert (shexp) |
Full Attention (L3, 7, 11, ...) | Attention Gates | Output Head (output.weight) |
Routers (gate_inp) |
File Size | Real-World Impact |
|---|---|---|---|---|---|---|---|---|---|
| Generic APEX Mini | IQ2_S (2.50 bpw) |
Q3_K (5 layers) |
Q4_K / Q3_K |
Q3_K |
Compressed | Q3_K_M |
Compressed | approx. 12.5 GB | Severe syntax drops, broken indentation, high perplexity in <think>. |
| MiniPlus V2.1 | IQ3_XXS + Q3_K |
Q3_K (20 layers) |
Q5_K |
Q4_K + Q6_K |
Q8_0 |
Q6_K |
F32 |
14.75 GB | Measured Q5_K_M tier. Fits 24GB GPUs effortlessly. Strict 0% 2-bit policy. |
| NanoPlus V2.1 (Current) | IQ3_XXS (down) + IQ2_S (gate) + IQ2_XXS (up) |
Q3_K (down) + IQ3_XXS (gate/up) |
Q4_K |
Q4_K (q/k/v) + Q6_K (attn_output) |
Q8_0 |
Q6_K |
F32 |
12.55 GB | Calibrated 12–13 GB tier. Leaves >3 GB free VRAM on 16GB cards for 32k context with zero AVX2 CPU stalls. |
[!TIP]
Deployment & System Architecture Guide
- Full GPU VRAM Offload (16GB+ VRAM,
-ngl 99): Effortless full offload with native 32K–64K context support on 16GB cards (RTX 4080 / RTX 4070 Ti Super), and native 256K context on 24GB workstations (RTX 3090 / 4090 / 5090).- System RAM Streaming Specialist (DDR4/DDR5 & Massive Context): Specially engineered to run either partially or entirely out of system RAM across large or full context windows. By utilizing linear SIMD-optimized
Q4_Kattention projections and preserving critical down-projections inIQ3_XXS, AVX2 CPU dequantization stalls are eliminated.Explore our official collection: APEX-I NanoPlus Abliterated MoE Collection.
Quality Spectrum: APEX-I-NanoPlus vs. Standard Flat Quantizations
How the handcrafted APEX-I-NanoPlus-V2.1 architecture compares against standard flat quantizations in llama.cpp on 35B Mixture-of-Experts architectures:
| Quantization Format | Bits Per Weight (BPW) | Model Footprint (Disk / VRAM) | Memory Headroom (16GB GPU) | Token Fidelity & Syntactic Stability Tier |
|---|---|---|---|---|
| FP16 / BF16 (Uncompressed) | 16.0 bpw | approx. 70.0 GB | OOM (Requires 80GB GPU) | 100% full uncompressed reference baseline. |
| Standard Q8_0 | 8.50 bpw | approx. 38 GB | OOM (Requires 48GB+ VRAM) | Virtually lossless; excessive memory overhead for consumer hardware. |
| Standard Q6_K | 6.56 bpw | approx. 30 GB | OOM (Requires 32GB+ VRAM) | Near-lossless FP16 fidelity; requires multi-GPU or 32GB+ VRAM setups. |
| Standard Q4_K_M | 4.50 bpw | approx. 20.0 GB | OOM on 16GB GPU | Standard industry trade-off; cannot fit in 16GB VRAM. |
| APEX-I-MiniPlus V2.1 Abliterated | 3.52 bpw | 14.75 GB (13.74 GiB) | Fits 16GB with tight context | Maximum fidelity near-lossless Q5_K / Q6_K tier. Full native 256K context on 24GB workstations. |
| APEX-I-NanoPlus-V2.1 Abliterated (CURRENT) | 2.90 bpw | 12.55 GB (11.69 GiB) | approx. 3.5 GB free VRAM | Solid Q4_K_M fidelity tier at only 12.55 GB (82.1% weight reduction). Enables full offload on 16GB GPUs with 32k context and zero AVX2 CPU stalls. |
| Standard Q3_K_M | 3.44 bpw | approx. 16.5 GB | OOM on 16GB GPU | Noticeable syntax drop, bracket corruption, and tokenizer classification noise. |
| Standard IQ2_S / Generic APEX Mini | 2.50 bpw | approx. 12.2 GB | Fits 16GB GPU | Severe reasoning breakdown, high perplexity spikes in <think> chains. |
Model Files & Technical Specifications
| File Name | File Size | Memory Footprint | BPW | Description |
|---|---|---|---|---|
Occamy-1.0.APEX-I-NanoPlus-V2.1-Abliterated.gguf |
12.55 GB (11.69 GiB) |
11.69 GiB |
2.90 BPW | Core uncensored frontier linear attention & multimodal reasoning MoE |
mmproj-Q8_0.gguf |
614.19 MB (585.74 MiB) |
585.74 MiB |
8.50 BPW | Dedicated Q8_0 multimodal vision projector for document & image reasoning |
occamy-1.0-mtp-Q8_0.gguf |
1.99 GB (1.85 GiB) | 1.85 GiB | 8.50 BPW | Official Accio MTP speculative draft head (high fidelity) |
occamy-1.0-mtp-Q4_K_M.gguf |
1.26 GB (1.17 GiB) | 1.17 GiB | 5.42 BPW | Official Accio MTP speculative draft head (low VRAM) |
tensor-map.tsv |
23.5 KB | — | — | Complete 733-tensor mapping audited directly from compiled weights |
chat_template.jinja |
10.5 KB | — | — | Hardened agentic chat template with multi-level reasoning control |
- Base Model: Accio-Lab/occamy-1.0
- Total Tensors: Exactly 733 tensors across 40 layers and 256 micro-experts
- Context Length: 262,144 tokens (native 256K)
Surgical Tensor Quantization Map (Audited from GGUF)
The exact tensor breakdown below has been verified directly from the compiled binary weights:
| Layer Group | Sub-Component / Tensor | Qty | Precision | Engineering Rationale |
|---|---|---|---|---|
| Global Output Head | output.weight |
1 | Q6_K |
Preserves near-FP16 token classification; eliminates syntax errors, bracket drops, and hallucinations. |
| Global Embeddings | token_embd.weight |
1 | Q4_K |
High-fidelity vocabulary embedding representation. |
| All Normalizations | output_norm, attn_*_norm, post_attention_norm, ssm_norm |
131 | F32 |
100% uncompressed numerical stability across all 40 layers. |
| Expert Routers | blk.*.ffn_gate_inp, ffn_gate_inp_shexp |
80 | F32 |
100% uncompressed routing fidelity across 256 micro-experts; zero router drift. |
| Attention Gates | blk.*.attn_gate.weight (30 Hybrid Layers) |
30 | Q8_0 |
High-precision attention gating across hybrid DeltaNet recurrence layers; eliminates crosstalk. |
| Shared Foundation Experts | blk.*.ffn_{gate,down,up}_shexp (All 40 Layers) |
120 | Q4_K |
Foundation knowledge backbone active on 100% of tokens; protected in linear Q4_K for fast streaming. |
| Periodic Full Attention | blk.{3,7,11,...}.attn_q/k/v (10 Anchor Layers) |
30 | Q4_K |
Full quadratic attention anchor checkpoints for deep needle-in-a-haystack retrieval. |
| Periodic Full Attention | blk.{3,7,11,...}.attn_output (10 Anchor Layers) |
10 | Q6_K |
High-precision attention output projection over deep context. |
| Recurrent SSM Scales | blk.*.ssm_alpha, ssm_a, ssm_conv1d, ssm_dt |
120 | F32 |
Guarded in uncompressed FP32 to prevent DeltaNet recurrent state drift. |
| Linear Attention & SSM | blk.*.attn_qkv, ssm_beta, ssm_out |
90 | Q4_K |
Linear AVX2 execution; zero SIMD CPU stalls during system RAM streaming. |
| Border MoE Down-Proj | Layers 0–1 & 38–39 (ffn_down_exps) |
4 | Q3_K |
Linear SIMD execution optimized for token entry and exit stability. |
| Border MoE Gate/Up | Layers 0–1 & 38–39 (ffn_gate/up_exps) |
8 | IQ3_XXS |
High-density boundary protection guided by imatrix. |
| Core MoE Down-Proj | Layers 2–37 (ffn_down_exps) |
36 | IQ3_XXS |
Fortified 3.06 bpw residual stream; preserves core mathematical, coding, and reasoning capacity. |
| Core MoE Gating | Layers 2–37 (ffn_gate_exps) |
36 | IQ2_S |
High-precision 2.50 bpw SwiGLU gating; eliminates activation noise. |
| Core MoE Up-Proj | Layers 2–15 (ffn_up_exps) |
14 | IQ2_S |
Enhanced 2.50 bpw precision for sensitive early-intermediate feature extraction. |
| Core MoE Up-Proj | Layers 16–37 (ffn_up_exps) |
22 | IQ2_XXS |
Extreme 2.06 bpw compression in deep MoE layers to reach exact 12.55 GB envelope. |
Inference Quickstart
1. llama-cli (Console Generation)
llama-cli \
-m Occamy-1.0.APEX-I-NanoPlus-V2.1-Abliterated.gguf \
--mmproj mmproj-Q8_0.gguf \
-p "<|im_start|>user\nHello! Explain your architecture.<|im_end|>\n<|im_start|>assistant\n" \
-ngl 99 -c 4096 --temp 0.6 --top-p 0.95
2. llama-server (OpenAI-Compatible API)
llama-server \
-m Occamy-1.0.APEX-I-NanoPlus-V2.1-Abliterated.gguf \
--mmproj mmproj-Q8_0.gguf \
--port 8080 \
-ngl 99 -c 16384
3. Speculative Decoding with MTP Draft Head (1.8x to 2.3x Speedup)
llama-server \
-m Occamy-1.0.APEX-I-NanoPlus-V2.1-Abliterated.gguf \
--draft-model occamy-1.0-mtp-Q4_K_M.gguf \
--spec-type draft-mtp \
--spec-draft-n-max 2 \
--mmproj mmproj-Q8_0.gguf \
--port 8080 \
-ngl 99 -c 16384
[!IMPORTANT]
CRITICAL ADVISORY FOR CODING WORKFLOWS: PREVENTING SYNTAX & TOKEN SWAPPING
In programming code, brackets (
{,}), assignment operators (=), and indentation whitespace repeat constantly across multi-line structures.Common Issue: Many local frontends (such as LM Studio defaults, Ollama, or web interfaces) ship with
repeat_penaltyset to1.1or1.15. While this prevents loops in creative prose, applying repeat penalties to code artificially penalizes necessary syntax tokens. When the logit of{drops, the model is forced to emit the next closest mathematical token (=or[), resulting in character swapping or dropped/doubled whitespace.Resolution:
- Disable Repeat Penalties (Required for Code):
repeat_penalty: 1.0(strictly disabled)presence_penalty: 0.0frequency_penalty: 0.0- Calibrate Samplers:
temperature: 0.60(or0.20-0.30for strict, deterministic code syntax)min_p: 0.05(prunes low-probability noise tokens effectively)top_p: 0.95top_k: 20- Native Jinja Formatting: Always pass the
--jinjaflag so the tokenizer handles leading-space BPE tokens cleanly.
[!TIP]
HARDENED AGENTIC CHAT TEMPLATE (JINJA)
An optimized
chat_template.jinjais included at the root of this repository. It hardens agent workflows and multi-turn stability:
- Native
reasoning_effortMulti-Level Control:
low/minimal: Keeps internal thinking concise and focused strictly on immediate execution steps to minimize latency in automated loops.medium(default): Balanced, structured reasoning process with standard analytical depth.high/xhigh: Guides the model to formulate a clear implementation plan upfront before generating code, avoiding circular self-doubt loops.none/off: Closes the thinking block immediately (<think>\n\n</think>) when reasoning is disabled.- Tool-Calling Safeguard (Anti-Premature Stop): Prevents the model from terminating a turn (
<|im_end|>) at a colon or action declaration prior to outputting<tool_call>.- Multi-Turn Thinking Memory: Preserves historical
<think>blocks across turns by default, preventing context distribution drift in 78K+ token runs.Usage with llama-server:
llama-server -m Occamy-1.0.APEX-I-NanoPlus-V2.1-Abliterated.gguf --chat-template-file chat_template.jinja --reasoning-effort medium
Community Compute Fund & Student Support
All IsValorum quantizations will always remain completely free and open to the public without paywalls.
However, cloud GPU compute is expensive and funded entirely out-of-pocket. If you find these handcrafted builds to be of high quality and they benefit your day-to-day workflow, you can support future Day-0 releases on Ko-fi. Every contribution directly funds the cloud compute hours required for new model runs and helps support my university education:
(When supporting on Ko-fi, feel free to leave a note with your Hugging Face handle and the specific model you would like prioritized).