Liquid AI ने LFM2.5-DSpark लॉन्च किया, जिससे अधिकतम 3.2 गुना तेज़ इनफ़रेंस मिलता है

Liquid AI ने 20 अगस्त 2026 को अपने LFM2.5 परिवार के तीन मॉडलों के लिए speculative-decoding ड्राफ्ट चेकपॉइंट जारी किए, जिसमें एकल H100 GPU पर अधिकतम 3.18x और Apple-silicon MacBook पर अधिकतम 2.87x थ्रूपुट वृद्धि का उल्लेख किया गया, जबकि मॉडल आउटपुट में कोई परिवर्तन नहीं हुआ। The LFM2.5-DSpark release covers drafters for LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and the mixture-of-experts LFM2.5-8B-A1B, each adding roughly 300 million parameters of draft overhead on top of the target model.

The checkpoints ship in Safetensors and GGUF formats with day-one support in llama.cpp and SGLang, both integrations contributed upstream to the official codebases. Because speculative decoding only emits tokens the target model has verified, the company states the generated text is identical to what the target would produce alone under greedy decoding, so benchmark accuracy is unchanged.

Liquid AI’s measurements, run at batch size 1 and temperature 0 across five datasets, put the mean speedup for LFM2.5-2.6B at 2.67x on an H100 (323 to 864 tokens per second) and 2.27x on an M4 Max MacBook Pro (61 to 139 tokens per second). The largest single result came from LFM2.5-8B-A1B on MATH500, where throughput on the H100 rose 3.18x, from 428 to 1,362 tokens per second. The company also reports that DSpark cut function-calling latency by 57% on average for LFM2.5-2.6B across multi-tool scenarios, the headline result for the on-device agentic workloads the LFM2.5 line is aimed at.

DSpark डिकोडिंग को कैसे तेज़ करता है

The decode phase of LLM inference is memory-bound: most of the latency comes from streaming weights from DRAM into on-chip memory rather than from computation itself, which is why inference economics have become the field’s central engineering problem. Speculative decoding attacks this by having a small draft model propose a block of candidate tokens, then verifying the whole block in one forward pass of the target model, spreading the cost of loading the weights across every token checked.

DSpark, introduced in a जुलाई 2026 paper by DeepSeek researchers and deployed in that company’s DeepSeek-V4 serving system, combines three components: a parallel backbone that produces hidden states for all draft tokens in a single pass, a lightweight sequential head that models dependencies between neighboring tokens to keep acceptance rates from decaying late in the block, and a confidence-scheduled verifier that prunes low-confidence suffixes when verifying them would cost more than it saves. In DeepSeek’s production deployment, the paper reports per-user generation speedups of 60 to 85% over the prior MTP-1 baseline at matched throughput.

Liquid AI’s drafters follow that recipe with a simplified attention-only design: five layers, a block size of nine draft tokens per step, and a Markov head over a 128,000-token vocabulary, per the LFM2.5-2.6B-DSpark model card. Each drafter was trained for 15 epochs on a mix of supervised fine-tuning, chat, code, and function-calling data, with the checkpoint selected by highest acceptance rate rather than lowest loss. The exactness guarantee does the quality work: “Speculative decoding is exact: the target verifies every proposed token, so greedy output equals the target alone,” the GGUF model card states, with per-response timings exposing how many draft tokens were proposed and accepted.

संख्याओं में LFM2.5-DSpark

• 3.18x — रिपोर्टेड सर्वोत्तम GPU गति वृद्धि (LFM2.5-8B-A1B, MATH500, H100: 428 → 1,362 tok/s)
• 2.87x — रिपोर्टेड सर्वोत्तम ऑन‑डिवाइस गति वृद्धि (LFM2.5-1.2B-Instruct, HumanEval, M4 Max: 136 → 389 tok/s)
• 2.67x / 2.27x — पाँच डेटासेट्स में LFM2.5-2.6B के लिए औसत H100 / M4 Max गति वृद्धि
• 57%: मल्टी‑टूल परिदृश्यों में LFM2.5-2.6B के लिए औसत फ़ंक्शन‑कॉलिंग लेटेंसी कमी
• 295.7M–327.7M (ड्राफ्ट मॉडल पैरामीटर, लक्ष्य मॉडल 1.2B से 8B के बीच)
• 4.81 of 10, ब्लॉक आकार 9 पर LFM2.5-2.6B के लिए प्रति चरण औसत स्वीकृत ड्राफ्ट टोकन

जहाँ रिपोर्टेड गति वृद्धि घटती है

Liquid AI की अपनी तालिकाएँ दिखाती हैं कि लाभ समान नहीं हैं, और कंपनी कारण बताती है। LFM2.5-8B-A1B के लिए, ऑन‑डिवाइस सुधार औसतन केवल 1.18x है, जबकि तीन मॉडलों में सबसे अधिक स्वीकृति दर है; यह अंतर कंपनी का मानना है कि वर्तमान mixture-of-experts कार्यान्वयन llama.cpp के Metal बैकएंड में और टोकन ब्लॉक को सत्यापित करने से विशेषज्ञों के बीच अतिरिक्त वज़न ट्रैफ़िक के कारण है। LFM2.5-1.2B-Instruct के लिए, स्वीकृति दर डेटासेट के अनुसार पर्याप्त बदलती है, जिससे गति वृद्धि 52% तक बदल सकती है, टेक्स्ट वितरण पर निर्भर करते हुए, H100 पर MT‑Bench पर 1.66x से लेकर MATH500 पर 2.56x तक।

All figures are vendor-reported from Liquid AI’s own harness: SGLang on one H100 80GB in BF16 for GPU numbers, llama.cpp with experimental Metal kernels on an M4 Max with FP16 GGUF weights for on-device numbers, capped at 256 output tokens. The SGLang path requires a build with DSpark support for LFM2 targets, and the llama.cpp path requires the corresponding build, so the speedups depend on those integrations rather than shipping in a stable release of either engine.

Liquid AI की ऑन‑डिवाइस पहल अब तक

The DSpark release is the third LFM2.5-family update in just over a week. On 12 अगस्त 2026, the company shipped LFM2.5-VL-3B, a vision-language model for the edge, and on 19 अगस्त 2026 it published quantization-aware distilled Q4_0 checkpoints for the family. The through-line is the same: the company says the 2.6B model’s DSpark speedup on a MacBook pushes interactivity beyond the throughput offered by most proprietary cloud models, which it puts at roughly 140 tokens per second.

All three drafters are available now on Hugging Face: LFM2.5-1.2B-Instruct-DSpark, LFM2.5-2.6B-DSpark, and LFM2.5-8B-A1B-DSpark, with GGUF builds alongside for llama.cpp deployments.