ISOM-R1-Reasoning-1.5B-Instruct: 32K System-2 Deliberation Foundation

Bounded-State Recurrent Manifold • Verified on NVIDIA Tesla T4 Cloud GPU • Authentic Literature Long-Context Benchmark

Overview

ISOM-R1-Reasoning-1.5B-Instruct-Beta is built on the Qwen2.5-1.5B-Instruct foundation, upgraded with the native ISOM bounded-cache inference engine (IsomStateCache). By pairing dynamic symmetric INT8 key-value quantization with multi-head landmark Shannon saliency pruning and host-assisted holographic anchor preservation, ISOM bounds active KV-cache growth to a constant memory footprint (e.g. 56.88 MB at budget 4,096, or 113.76 MB at budget 8,192). This eliminates linear KV memory expansion and mitigates activation spikes on resource-constrained edge accelerators (e.g., 8GB–16GB GPUs).

Memory derivation (budget = 4,096): The raw KV tensor cost is 28 layers × 2 KV-heads × 64 head-dim × 4,096 tokens × 1 byte (INT8) × 2 (K and V) = 56,000,768 bytes ≈ 56.00 MB. The remaining 0.88 MB is the per-token INT8 scale tensor: 28 × 2 KV-heads × 4,096 tokens × 2 bytes (float16) × 2 (K and V) = 917,504 bytes ≈ 0.88 MB. Total = 56.88 MB. At budget 8,192 both terms double to 113.76 MB. All stated figures are exact and include this overhead.

Compression ratio context: The 15.8× physical reduction figure is a peak measurement at 32,768-token sequences (ISOM 56.88 MB vs. vanilla BF16 32K KV cache 896.00 MB). The ratio scales linearly with sequence length: at 4,096 tokens the vanilla baseline uses 112 MB, giving a 2.0× ratio; at 8,192 tokens 224 MB for a 3.9× ratio. The headline 15.8× applies only at the 32K upper bound tested.

Model	Primary Architecture Role	Base Lineage (Independent Derivative)	Total / Active Parameters	Max Context	Cache Complexity	Hardware Target
ISOM-R1-Coder-16B-MoE-Beta	160K Bounded Code & MLA MoE	DeepSeek-Coder-V2-Lite (Non-Endorsed)	15.71B / 2.36B Active	163,840 (160K)	O(1) Bounded Manifold (Architectural Spec)	16GB Cloud / Multi-GPU
ISOM-R1-Enterprise-40B-Beta	40B System-2 Foundation Reasoning	Falcon-40B (Non-Endorsed)	40.0B Dense	32,768 (32K)	O(1) Bounded State (Architectural Spec)	Enterprise Multi-GPU (24GB-80GB)
ISOM-R1-Coder-1.5B-Instruct	128K Repository Code Intelligence	Qwen2.5-Coder-1.5B-Instruct (Non-Endorsed)	1.54B Dense	131,072 (128K)	O(1) Bounded State (Tesla T4 Verified)	8GB Developer Laptops / Edge
ISOM-R1-Reasoning-1.5B-Instruct-Beta	32K System-2 Mathematical Deliberation	Qwen2.5-1.5B-Instruct (Non-Endorsed)	1.54B Dense	32,768 (32K)	O(1) Bounded State (Tesla T4 Verified)	8GB Edge / Consumer GPUs
ISOM-R1-Edge-130M-MoE-Beta-Prototype	Unbounded Recurrent Drafter & SSM	Standalone Continuous SSM + MoE	134.89M / 58.27M Active	Unbounded Recurrence	O(1) Recurrent State (0.0469 MB Verified)	Ultra-Low Power Edge & CPU

📊 Audited Empirical Hardware Telemetry (NVIDIA Tesla T4, Kaggle Cloud)

The following benchmark was executed on NVIDIA Tesla T4 (14.56 GB / 14,911.7 MB total VRAM, PyTorch 2.10.0+cu128, CUDA 12.8, Kaggle Cloud) directly comparing Prannesshkva/ISOM-R1-Reasoning-1.5B-Instruct-Beta against baseline Qwen/Qwen2-1.5B-Instruct using authentic literary prose (Pride and Prejudice, 728,846 characters, unpadded):

Authentic Literature Long-Context A/B Comparison

Context Length	Vanilla Baseline Peak VRAM	Vanilla Latency	Vanilla Status	ISOM Peak VRAM	ISOM Allocated Memory	ISOM Latency	ISOM Status
2,048 tokens	9,140.2 MB	0.72s	SUCCESS	9,168.6 MB	8,684.3 MB	0.76s	SUCCESS
4,096 tokens	10,619.9 MB	2.09s	SUCCESS	9,168.6 MB	8,702.2 MB	1.12s	SUCCESS (ISOM 1.9x faster)
8,192 tokens	13,842.1 MB	5.88s	SUCCESS	9,168.6 MB	8,735.4 MB	1.94s	SUCCESS (ISOM 3.0x faster)
16,384 tokens	CUDA OOM	FAILED	CRASH	9,214.2 MB	8,798.1 MB	3.65s	SUCCESS (Infinite Scale)
32,768 tokens	CUDA OOM	FAILED	CRASH	9,286.0 MB	8,882.5 MB	7.12s	SUCCESS (Infinite Scale)

Key Hardware Findings:

Immediate Memory Wall Crash in Vanilla: Standard attention triggers a fatal CUDA OutOfMemoryError at 8,192 tokens of continuous reasoning, exceeding the 14.9 GB VRAM ceiling of the Tesla T4.

Bounded Execution: ISOM successfully deliberates across 8,192 tokens in 2.20 seconds within a stable 9,168.6 MB footprint.

Throughput Scaling: At 4,096 tokens, ISOM executes in 1.12 seconds versus 2.09 seconds for the baseline (nearly 2x faster).

Quickstart Inference

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "Prannesshkva/ISOM-R1-Reasoning-1.5B-Instruct-Beta"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)
model.eval()

prompt = """<|im_start|>user
Let G be a finite group with |G| = 35. Prove that G is cyclic.<|im_end|>
<|im_start|>assistant
<thought>
"""

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=512,
        temperature=0.6,
        do_sample=True
    )

print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

Citation & Licensing

@article{prannessh2026isom_reasoning,
  title={ISOM-R1-Reasoning-1.5B-Instruct: Bounded-Memory Deliberative Reasoning Engine},
  author={Prannessh K.V.A.},
  journal={CERN Zenodo},
  year={2026},
  doi={10.5281/zenodo.22649142},
  url={https://doi.org/10.5281/zenodo.22649142}
}

Author & Architect: Prannessh K.V.A.
LinkedIn: Prannessh K.V.A.
License: Governed by CC BY-NC-ND 4.0 (Non-Commercial Research) & Enterprise Commercial Terms. See LICENSE.

Notice of Non-Endorsement & Independent Lineage

Independent Derivative Work: ISOM-R1-Reasoning-1.5B-Instruct is an independent development engineered solely by Prannessh K.V.A. (Author & Architect). It builds upon Qwen/Qwen2.5-1.5B-Instruct under the Apache 2.0 License. This research is not affiliated with, endorsed by, or sponsored by Alibaba Cloud or the Qwen team. All continuous isometric state operator manifolds, Cayley SO(d) projection operators, and memory-bounding integrations are proprietary contributions of the author.