IsValorum/Cyber-Tiel-Coder-35B-A3B-APEX-I-MiniPlus-V2.1-GGUF

NEW: QWEN3.8 27B DENSE — VAL-APEX-I

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.

Independently computed on private cloud clusters. If this handcrafted release saves you VRAM and runs faster on your GPU, consider fueling the community compute fund on Ko-fi.

Explore the new canonical V3 releases. MiniPlus prioritizes Q5-class to near-Q6 fidelity without a fixed file-size ceiling; NanoPlus targets Q4-class quality at a practical footprint.

• Qwen3.8 27B — MiniPlus V3 — 14.16 GB (+13.21% improvement in WikiText-2 PPL vs GSQ-RCO IQ3_S).

• Qwen3.8 27B — NanoPlus V3 — 11.35 GB (+11.49% improvement in WikiText-2 PPL vs GSQ-RCO IQ3_S).

• Swift-1.5 Qwen3.8 27B Uncensored MTP — MiniPlus V3 — 14.28 GB (+5.70% improvement in WikiText-2 PPL vs GSQ-RCO IQ3_S).

• Swift-1.5 Qwen3.8 27B Uncensored MTP — NanoPlus V3 — 11.08 GB (+2.26% improvement in WikiText-2 PPL vs GSQ-RCO IQ3_S).

Percentages compare the reported WikiText-2 PPL with GSQ-RCO IQ3_S (7.07). The linked 27B cards document the evaluation protocol and model variants.

Explore the APEX-I MiniPlus V2.1 collection

EXPLORE THE ESTABLISHED 35B MoE MINIPLUS LINEUP

These are complementary APEX-I-MiniPlus V2.1 releases, not alternate downloads of the same model. Each receives the same tensor-by-tensor approach, companion MTP where supported, and a design suitable for full or partial system-RAM inference. Choose the model whose native strengths best fit the work you want to do:

• Tiel-Coder-35B-A3B APEX-I-MiniPlus-V2.1 — specialist agentic coding MoE tuned for high-speed local development and SWE-bench tasks.

• Qwen3.6-35B-A3B MTP APEX-I-MiniPlus-V2.1 — a versatile frontier MoE for broad reasoning, multilingual work, agents, tool use, and multimodal tasks.

• Qwen3.6-35B-A3B MTP APEX-I-MiniPlus-V2.1 Abliterated — the V2.1 refusal-ablated Qwen3.6 edition for users who deliberately prefer reduced refusal behavior.

• Ornith 1.5 APEX-I-MiniPlus-V2.1 — a software-engineering-focused MoE designed for repository-scale coding and autonomous engineering agents.

These remain distinct model families and editions with their own behavior and empirical results. Pick by workload and intended alignment behavior rather than treating them as interchangeable quantization variants.

Quick Navigation Index

1. Optimization History & Transparency Notice

2. Independent Community Benchmarks (Shortbench / NanoBench #03 Global) — includes the Known Limitation: Type/Schema Preservation Under Compaction.

3. Empirical Benchmarks & Fidelity Verification

4. Quality Spectrum: APEX-I-MiniPlus V2.1 vs. Standard Flat Quantizations

5. Model Files & Technical Specifications

6. Surgical Tensor Quantization Map (Audited from GGUF)

7. Hardware Throughput & Offload Benchmarks (RTX 30 / 40 / 50 & RAM Streaming)

8. The 24GB Miracle: Full 256K Context Runs In VRAM!

9. Recommended Configuration & Setup

10. Recommended Generation Parameters (Cyber Tiel Coder Official)

11. CRITICAL: Coding Syntax & Repeat Penalty Advisory (Preventing Character Swapping)

12. Hardened Agentic Chat Template & Reasoning Effort

13. Optional Support

Cyber-Tiel-Coder-35B-A3B APEX-I-MiniPlus-V2.1 GGUF

The Definitive Frontier MoE · Efficient System RAM Offload · Full 256K Context on 24GB Workstations

THE DEFINITIVE SPECIFICATION IN THE 14–15 GB CEILING

This APEX-I-MiniPlus-V2.1 release represents the specialized tensor-by-tensor configuration for sparse Mixture-of-Experts quantization within a 14–15 GB envelope. Every tensor across its 40 layers, 256 micro-experts, and decoupled companion MTP architecture has been mathematically allocated to maximize reasoning precision, preserve routing behavior, and prevent avoidable CPU dequantization stalls.

🏆 EMPIRICAL BENCHMARK & QUALITY COMPARISON

| Quantization Specification | File Size (Disk) | Memory Footprint (RAM/VRAM) | Average BPW | WikiText-2 Perplexity | Quality Tier Equivalent |

| :--- | :---: | :---: | :---: | :---: | :---: |

| Unquantized BF16 Base | 71.05 GB | 66.18 GiB | 16.00 BPW | approx. 7.46 (Reference) | Full precision baseline |

| APEX-I-MiniPlus V2.1 (CURRENT) | 14.75 GB | 13.74 GiB | 3.43 BPW | 7.5124 ± 0.20729 (ΔPPL +0.0524 / +0.70%) | Q5_K_L tier (bordering Q6_K) |

| APEX-I-NanoPlus | 12.55 GB | 11.69 GiB | approx. 2.93 BPW | 8.2849 ± 0.23438 (ΔPPL +0.8249 / +11.06%) | Solid Q4_K_M / Q4_K_L tier |

Prefer the ultra-compact Cyber-Tiel-Coder edition? APEX-I-NanoPlus compresses the same Cyber-Tiel-Coder family down to 12.55 GB (approx. 2.93 BPW) for agile 16GB VRAM & RAM streaming at Solid Q4_K_M / Q4_K_L tier quality, while MiniPlus V2.1 retains the higher 14.75 GB (3.43 BPW) Q5_K_L tier fidelity.

Routing: all recipe-designated gate_inp and gate_shexp tensors remain in uncompressed F32, preserving zero routing drift.

• Q6_K-Bordering Tier in Language Fidelity: WikiText-2 perplexity delta is exceptionally low (under +0.053 / < 0.71% vs. BF16), placing overall language representation at the boundary of a 28 GB Q6_K build within an agile approx. 14.75 GB footprint.

• Q5_K_L Tier in MoE Foundation Knowledge: All 120 shared expert tensors (shexp) across all 40 layers run in uncompressed Q5_K, keeping coding and reasoning domain knowledge intact across 100% of tokens.

• Solid 4-to-5-bit Routing & Attention Armor: Full quadratic attention in Q4_K and output head in Q6_K prevent token drift and formatting collapse.

DO NOT CONFUSE APEX-I-MINIPLUS WITH GENERIC COMMUNITY APEX-I-MINI!

Regardless of release version, NEVER confuse handcrafted APEX-I-MiniPlus builds with generic community APEX-I-Mini releases:

• Generic Community APEX-I-Mini: Uniformly compresses all core MoE experts down to aggressive 2-bit IQ2_S, leaves the sensitive token output head unarmored at 3-bit Q3_K_M, and compresses attention projections down to Q3_K. In deep reasoning models, this triggers severe perplexity spikes, syntax errors, and broken code brackets.

• Handcrafted APEX-I-MiniPlus V2.1: Applies a custom tensor-by-tensor architecture that preserves specified router gates in uncompressed F32, armors the token output head in high-precision Q6_K, safeguards attention gates in Q8_0, and keeps core reasoning experts at calibrated 3-bit treatment.

SYSTEM RAM INFERENCE: FULL OR PARTIAL

This APEX-I-MiniPlus release is designed for full or partial system-RAM inference. Depending on the processor, memory bandwidth, and DDR4/DDR5 configuration, generation can range from 20 to 45 tok/s. With partial GPU offload, systems that cannot fit 128K or more context entirely in VRAM can place the remaining model and context load in system RAM, maintaining stable, responsive generation at longer context lengths.

---

Optimization History & Transparency Notice

We maintain our previous releases publicly as a transparent engineering record of continuous optimization. Below is the exact evolutionary roadmap of our MiniPlus architectures:

| Specification | Core Experts (10–29) | Edge Experts (0–9, 30–39) | Shared Expert (shexp) | Full Attention (L3, 7, 11, ...) | Attention Gates (30 Layers) | Output Head (output.weight) | Routers (gate_inp) | Size / Overhead | Real-World Impact |

| :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :--- |

| Generic APEX Mini | IQ2_S (2.50 bpw) | Q3_K (only 5 layers) | Q4_K / Q3_K | Q3_K | Compressed | Q3_K_M | Compressed | Baseline (approx. 12.5 GB) | Severe syntax errors, broken code indentation, high perplexity in <think>. |

| MiniPlus V2.1 (CURRENT) | IQ3_XXS | Q3_K (20 layers) | Q5_K (All 40 layers) | Q4_K (q/k/v) + Q6_K (output) | Q8_0 | Q6_K | F32 | Definitive Build (approx. 13.74 GiB) | Zero AVX2 CPU stalls and efficient streaming when offloading bulk of the model to system RAM (DDR4/DDR5). Clean decoupled companion MTP draft head provided. |

Deployment & System Architecture Guide

• Full GPU VRAM Offload (24GB+ VRAM, -ngl 99): Effortless full offload with native 256K context support. Blistering throughput on RTX 3090 / 4090 / 5090 GPUs.

• System RAM Streaming Specialist (DDR4/DDR5 & Massive Context): Specially engineered to run either partially or entirely out of system RAM across large or full (+160k to 256k) context windows. By replacing non-linear codebooks with linear SIMD-optimized Q3_K edge experts and upgrading shared foundation experts to Q5_K across all 40 layers, AVX2 CPU dequantization stalls are completely eliminated. Depending on your processor architecture and memory bandwidth (dual-channel DDR4 or high-speed DDR5 6000+ MT/s), streaming generation in system RAM can approach speeds remarkably close to full VRAM execution, allowing both the dedicated Q8_0 multimodal vision projector (mmproj) and optional companion Multi-Token Prediction draft model to be used by compatible runtimes for speculative decoding, while the main model weights stream effortlessly from system RAM.

Explore the complete family of APEX-I-MiniPlus models in our official collection:

APEX-I-MiniPlus V2.1 Hub Collection.

---

---

🏅 Independent Community Benchmarks (Shortbench / NanoBench v0.5.0)

Verified Autonomous Agent Evaluation: Evaluated on the Shortbench / NanoBench v0.5.0 autonomous coding and agentic benchmark (12 objective software engineering, pipeline, and system integrity tasks).

In extensive independent testing conducted by community member @derFusel on a consumer NVIDIA GeForce RTX 4070 (12 GB VRAM) workstation, Cyber-Tiel-Coder-35B-A3B-APEX-I-MiniPlus-V2.1 achieved a perfect first-pass resolution score in both non-compacted runs (reasoning_effort=high, q8_0 KV cache); the 96K run ranks #03 globally and the 160K run #05:

Benchmark Run	Total Score	Solved Tasks	Completion Time	Global Ranking	Tested Hardware & Configuration	Public Verification Link
Cyber-Tiel-Coder (96K Context)	100.0%	12 / 12	16.22 min	#03 Global	RTX 4070 (12GB) / i7-14700K / 32GB RAM (Pi Agent, --n-cpu-moe 16)	Verified Run 775f8eb7
Cyber-Tiel-Coder (160K Context)	100.0%	12 / 12	20.68 min	#05 Global	RTX 4070 (12GB) / i7-14700K / 32GB RAM (Pi Agent, --n-cpu-moe 19)	Verified Run 32804f02
Cyber-Tiel-Coder (Compacted)	98.3%	10 / 12	16.70 min	#18 Global	RTX 4070 (12GB) / i7-14700K / 32GB RAM (Pi Agent, context compaction)	Verified Run 2d46ce8b

Key Empirical Takeaways:

• Outperforming Frontier Cloud Models on Local Hardware: Solved all 12 tasks in 16.22 minutes on a single consumer 12 GB GPU, outpacing commercial frontier API runs such as Gemini 3.8 Flash (17.10 min) on the same test suite.
• Flawless Multi-Step Code Generation (non-compacted runs): Zero syntax errors, zero tool loop-guards, and first-pass execution across asynchronous file systems, date normalizers, and transactional state recovery tasks.
• Verified llama-server Parameters for 12 GB GPUs: High-throughput configuration reported by tester: -ngl 999 -ctk q8_0 -ctv q8_0 -b 8192 -ub 2048 -n 8192 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.05 --jinja --cache-reuse 256 -fa on --cont-batching --n-cpu-moe 16 -c 98304 --load-mode dio.

Known Limitation: Type/Schema Preservation Under Compaction

The two 100% runs were obtained with reasoning_effort=high, q8_0 KV cache, and no context compaction. In the compacted run (98.3%, 10/12), only two verifier groups were lost: string_id in coding-05 (15 pts, dropping it to 85/100) and mixed_schema in systems-02 (5 pts, dropping it to 95/100). Every retry and vendor-consistency check in coding-05 still passed. This points to type/schema preservation under reduced reasoning effort and/or history compaction rather than a retry-logic, context-size, KV-cache, or APEX weight-quantization defect.

For schema-sensitive agentic tasks, use reasoning_effort=high and keep context compaction disabled (or raise its threshold). If compaction is required, keep the following invariant in the system prompt outside the compacted region: "Preserve exact field types and schema. String IDs must remain strings even if numeric-looking; never coerce identifiers. Preserve mixed/heterogeneous schemas; do not normalize or unify field types. Verify these invariants before finishing."

---

Supplementary Evaluation: SWE-bench LLMBench Multi-File Bug Fixing (Occamy V2 Reference)

Historical Architecture Baseline: zephel01 independently benchmarked Occamy V2 on an NVIDIA RTX 5090 (32GB) across 348 unseeded trials on SWE-bench multi-file Python bug-fixing tasks (llmbench), resolving 93.3% at 32K and 90.0% at 65K context with only +0.5 GB VRAM expansion.

---

Empirical Benchmarks & Fidelity Verification

The comparison table near the top consolidates the model-specific BF16 baseline, final GGUF PPL, delta, published main-file size, BPW, and fidelity tier. The routing treatment is preserved in the note directly beneath it.

Quality Spectrum: APEX-I-MiniPlus V2.1 vs. Standard Flat Quantizations

How the handcrafted APEX-I-MiniPlus 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) | Perplexity Delta (vs. FP16 Baseline) | Token Fidelity & Syntactic Stability Tier |

| :--- | :---: | :---: | :---: | :--- |

| FP16 / BF16 (Uncompressed) | 16.0 bpw | 71.05 GB | 0.00 (Reference) | 100% full uncompressed reference fidelity. |

| Standard Q8_0 | 8.50 bpw | approx. 38 GB | approx. +0.01 | Virtually lossless; excessive memory overhead for consumer hardware. |

| Standard Q6_K | 6.56 bpw | approx. 30 GB | approx. +0.02 to +0.05 | Near-lossless FP16 fidelity; requires multi-GPU or 32GB+ VRAM setups. |

| 🏆 APEX-I-MiniPlus V2.1 (IsValorum) | 3.43 bpw | 14.75 GB (13.74 GiB) | ΔPPL +0.0524 (+0.70%) | Measured GGUF result: 7.5124 ± 0.20729 versus the approx. 7.46 BF16 baseline, supporting a Q5_K_L-class fidelity tier (bordering Q6_K) at less-than-Q3_K_M weight, at a 78.6% weight-size reduction. Full native 256K context on standard 24GB workstations. |

| Standard Q5_K_M | 5.50 bpw | approx. 25 GB | approx. +0.05 to +0.10 | Commercial transparent threshold; exceeds standard single 24GB GPU limits. |

| Standard Q4_K_M | 4.50 bpw | approx. 20 GB | approx. +0.15 to +0.25 | Standard industry trade-off; requires context offload compromises. |

| Standard Q3_K_M / Q3_K_S | 3.44 bpw | 16.7 GB | approx. +0.40 to +0.85 | Noticeable syntax drop, bracket corruption, and tokenizer classification noise. |

| Standard IQ2_S / Generic APEX Mini | 2.50 bpw | approx. 12.5 GB | approx. +1.50 to +3.00+ | Severe reasoning breakdown, high perplexity spikes in <think> chains. |

Model Files & Technical Specifications

| File Name | File Size | Memory Footprint | BPW | Description |

| :--- | :--- | :--- | :--- | :--- |

| Cyber-Tiel-Coder-35B-A3B.APEX-I-MiniPlus-V2.1.gguf | 14.75 GB (13.74 GiB) | 13.74 GiB | 3.43 BPW | Uncensored agentic coding, SWE-bench vulnerability analysis, reasoning & multimodal MoE |

| mmproj-Q8_0.gguf | 614.39 MB (585.93 MiB) | 585.93 MiB | 8.50 BPW | Dedicated Q8_0 multimodal vision projector for document & code architecture reasoning |

| mtp-Cyber-Tiel-Coder-35B-A3B.gguf | 1.48 GB (1.38 GiB) | 1.35 GiB | — | Companion Multi-Token Prediction (MTP) draft head for optional speculative decoding acceleration |

| Complete download | 16.84 GB (15.69 GiB) | 15.69 GiB | — | Main GGUF base + Q8_0 vision projector + companion MTP draft head |

• Base Model: Cyber-Tiel-Coder-35B-A3B-GGUF-MTP (derived from huihui-ai/Huihui-Ornith-1.5-35B-A3B-abliterated)

• Parameters: 35.2B total (approx. 2.6B to 3.2B active per token)

• Architecture: 40 layers, 256 micro-experts (8 active per token) + hybrid linear attention / DeltaNet recurrent layers

• 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 | Q5_K | Foundation knowledge backbone active on 100% of tokens; protected in high-precision linear Q5_K. |

| 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 | Armored 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 | Q3_K | Linear AVX2 execution; zero SIMD CPU stalls during system RAM streaming. |

| Edge MoE Experts | Layers 0–9 & 30–39 (ffn_*_exps) | 60 | Q3_K | Linear SIMD execution optimized for system RAM offload. |

| Core MoE Experts | Layers 10–29 (ffn_*_exps) | 60 | IQ3_XXS | Calibrated with importance matrix (imatrix) for maximum compactness in deep layers. |

---

Hardware Throughput & Offload Benchmarks (RTX 30 / 40 / 50 & RAM Streaming)

Empirically verified in Unsloth Studio & llama.cpp:

| Hardware Target | Offload Mode | Generation Speed (Est.) | Prompt Prefill Speed (Est.) | Highlights |

| :--- | :--- | :---: | :---: | :--- |

| NVIDIA RTX 5080 / 5090 (Blackwell) | Full GPU (-ngl 99) | approx. 247 – 251 tok/s | 2,800 – 3,900+ tok/s | Empirically verified on RTX 5090 by zephel01 (Occamy V2 Reference) |

| NVIDIA RTX 4090 (24GB GDDR6X) | Full GPU (-ngl 99) | 90 – 115+ tok/s | 2,000 – 2,800+ tok/s | Linear attention layers slash prefill latency |

| NVIDIA RTX 3090 (24GB GDDR6) | Full GPU (-ngl 99) | 72 – 88+ tok/s | 1,500 – 2,200+ tok/s | Full 256k native window in VRAM |

| Workstation / Laptop (DDR4 / DDR5 RAM) | Hybrid Offload (Few layers in VRAM) | Hardware-dependent | Hardware-dependent | Zero AVX2 CPU stalls; efficient streaming from system RAM |

• Aggressive Hybrid Offload Profile: Hybrid offload supports reasoning-enabled generation with limited VRAM while the remaining model weights stream from system RAM.

Empirical Testbed Architecture & Desktop/Server Scaling

• Empirical Benchmark Hardware: The hybrid offload and system RAM streaming behavior documented above was measured on a consumer laptop powered by an Intel 12th Gen Alder Lake architecture featuring a hybrid design of Performance Cores (P-Cores) and Efficient Cores (E-Cores) paired with dual-channel system RAM and constrained laptop power/thermal envelopes.

• Thread Scheduling & E-Core Contention: In hybrid architectures like Alder Lake, OS thread scheduling across background E-Cores and lower single-core mobile power limits introduce memory bandwidth and thread synchronization overhead during CPU dequantization.

• Dramatic Scaling on Higher-End Processors: When running on desktop or server processors (such as modern AMD Ryzen 7000 / 9000 Zen 4/5 series or high-TDP Intel desktop platforms with dedicated performance cores, large L3 caches, and high-bandwidth dual- or quad-channel DDR5 running at 6000+ MT/s), streaming generation speeds and prefill throughput will scale dramatically higher, substantially exceeding these measured mobile numbers.

---

The 24GB Miracle: Full 256K Context Runs In VRAM!

Cyber-Tiel-Coder-35B-A3B APEX-I-MiniPlus-V2.1 fits the entire 256K context window within 24GB VRAM:

| Context Length | Model Weights (Est.) | KV Cache (q8_0, 4 slots) | Compute Buffers | Total GPU VRAM (Est.) | Feasibility |

| :--- | :--- | :--- | :--- | :--- | :--- |

| 32,768 (32k) | 13.74 GiB | 0.58 GiB | 1.80 GiB | 16.56 GiB | Full offload on 24GB; partial on 16GB |

| 65,536 (64k) | 13.74 GiB | 0.92 GiB | 1.95 GiB | 17.05 GiB | Effortless fit on 24GB GPUs |

| 131,072 (128k)| 13.74 GiB | 1.58 GiB | 2.22 GiB | 17.98 GiB | Effortless fit on 24GB GPUs |

| 262,144 (256k)| 13.74 GiB | 2.92 GiB | 2.80 GiB | 19.90 GiB | FULL 256K NATIVE IN VRAM! |

Note: Leaves comfortable headroom for display drivers and compute buffers on standard 24GB GPUs (RTX 3090, RTX 4090, RTX 5090).

💡 Empirical 16GB GPU Verification (Single Stream / Desktop)

While theoretical multi-slot server buffers estimate approx. 16.5 GiB, independent hardware testing by zephel01 on an RTX 5090 (Occamy V2 Reference) confirmed that single-stream desktop inference consumes only 14.6 GB at 32,768 ctx and only 15.1 GB at 65,536 ctx (-ctk q8_0 -ctv q8_0 -fa on). This empirically proves that full 65K context runs completely in VRAM on 16GB cards (RTX 4080 / RTX 5080) without system RAM offload!

---

Recommended Configuration & Setup

llama-server.exe   -m Cyber-Tiel-Coder-35B-A3B.APEX-I-MiniPlus-V2.1.gguf   --mmproj mmproj-Q8_0.gguf   -md mtp-Cyber-Tiel-Coder-35B-A3B.gguf   --spec-type draft-mtp   --jinja   -ngl 99   --ctx-size 262144   --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0   --port 8080

Standard execution (without companion MTP):

llama-server.exe   -m Cyber-Tiel-Coder-35B-A3B.APEX-I-MiniPlus-V2.1.gguf   --mmproj mmproj-Q8_0.gguf   --jinja   -ngl 99   --ctx-size 262144   --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0   --port 8080

Coding note: The upstream sampling metadata uses --min-p 0, while the verified coding runs and the advisory below use --min-p 0.05. Use 0.05 for coding/agentic workloads.

⚙️ Recommended Generation Parameters (Cyber Tiel Coder Official)

Sampling metadata recorded from Cyber-Tiel-Coder-35B-A3B-GGUF-MTP in the completed build:

| Hyperparameter | Value | Description / Creator Notice |

| :--- | :---: | :--- |

| Temperature | 0.60 | Recommended official default for code generation & reasoning (--temp 0.6). |

| Top-P | 0.95 | Source GGUF sampling metadata (--top-p 0.95). |

| Top-K | 20 | Source GGUF sampling metadata (--top-k 20). |

| Min-P | 0.00 upstream / 0.05 coding | Source GGUF metadata is 0.00; the coding advisory and verified 100% runs use 0.05 to prune low-probability syntax noise. |

| Template Engine | --jinja | Recommended official chat template flag. |

| Reasoning Effort (Agentic) | high | Recommended for schema-sensitive agentic tasks; low/minimal may normalize field types under compaction. See Known Limitation. |

| Context Size | 262144 | Full native 256K context support (--ctx-size 262144). |

| Max New Tokens | Runtime-dependent | Select for target workload. |

🔍 Model Inherent Behavior vs. Quantization Fidelity Notice

Any behavioral nuances, stylistic tendencies, domain-specific habits, or zero-shot edge-case oversights stem entirely from the original unquantized checkpoint weights and fine-tuning distribution, NOT from the APEX-I quantization process.

Handcrafted APEX-I-MiniPlus strictly preserves mathematical tensor fidelity—keeping 100% of expert routing matrices (gate_inp) in uncompressed F32 (zero router drift), armoring the token output head in Q6_K, and safeguarding attention gates in Q8_0. Empirical verification records the final GGUF perplexity at 7.5124 ± 0.20729, a ΔPPL +0.0524 (+0.70%) versus the approx. 7.46 BF16 baseline.

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_penalty set to 1.1 or 1.15. 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.

Verified Upstream Behavior: This self-correcting behavior (where the model notices the mistake in its thinking loop but repeats the substitution) is documented on official upstream base checkpoints and Q8 builds (Ornith Discussion #34 and Discussion #22). It is completely eliminated by proper sampling configuration:

1. Disable Repeat Penalties (Required for Code):

• repeat_penalty: 1.0 (strictly disabled)

• presence_penalty: 0.0

• frequency_penalty: 0.0

2. Calibrate Samplers:

• temperature: 0.60 (or 0.20 - 0.30 for strict, deterministic code syntax)

• min_p: 0.05 (prunes low-probability noise tokens effectively)

• top_p: 0.95

• top_k: 20

3. Native Jinja Formatting: Always pass the --jinja flag so the tokenizer handles leading-space BPE tokens ( { vs {, = vs =) cleanly.

HARDENED AGENTIC CHAT TEMPLATE (JINJA)

An optimized chat_template.jinja is included at the root of this repository. It hardens agent workflows and multi-turn stability:

1. Native reasoning_effort Multi-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.

2. 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>.

3. 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 Model.gguf --chat-template-file chat_template.jinja --reasoning-effort medium

7. 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).