microsoft · text

harrier-oss-v1-270m

microsoft/harrier-oss-v1-270m

harrier-oss-v1-270m at Q4_K_M is exactly 253,114,048 bytes (0.24 GiB / 0.25 GB) — an effective 7.553 bits per weight, not the nominal 4. Its KV cache at 32K is 0.56 GiB.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
268M
Architecture
gemma3
18 layers
Context
32,768
native (config.json)
License
mit

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
Q3_K_L0.23 GiB246,386,3687.352—mykor
IQ4_XS0.23 GiB249,555,2327.447—cstr
Q4_K0.23 GiB250,661,1527.480—cstr
Q4_K_M0.24 GiB253,114,0487.553—mykor
Q5_K_M0.24 GiB260,026,0487.759—mykor
Q5_K0.25 GiB263,194,9127.854—cstr
Q6_K0.26 GiB282,973,8888.444—mykor
Q8_00.27 GiB291,544,7688.700—mykor
Q8_00.28 GiB300,796,1928.976—cstr
BF160.51 GiB542,834,36816.198—mykor

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0960.07 GiB0.07 GiB—18 / 0 / 0
8,1920.14 GiB0.14 GiB—18 / 0 / 0
16,3840.28 GiB0.28 GiB—18 / 0 / 0
32,7680.56 GiB0.56 GiB—18 / 0 / 0
65,5361.13 GiB1.13 GiB—18 / 0 / 0
131,0722.25 GiB2.25 GiB—18 / 0 / 0

Compare with

same modality, comparable size
vs Llama-3.2-1B-Instructvs ced-basevs gemma-3-1b-itvs Qwen3-1.7Bvs embeddinggemma-300mvs Qwen3-0.6BAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 0.14 GiB. The real file is 0.24 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
18
Attention heads
4
KV heads
1
Head dim
256
Hidden size
640
Vocab
262,144
Sliding window
512
SWA period
1
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does harrier-oss-v1-270m need?
Q4_K_M is exactly 253,114,048 bytes (0.24 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is harrier-oss-v1-270m's KV cache?
0.56 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Which quantization of harrier-oss-v1-270m should I use?
Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.