Precision & Context Tiers — ML Junction docs

Quantization describes the serving route, not the abstract model family. Different providers may serve the same model at different floating-point or integer precision. Do not infer precision from the model name.

Value · Meaning · Practical interpretation
fp16 · IEEE 16-bit floating point · Common full/near-full serving precision
bf16 · Brain floating point 16 · Wide exponent range with reduced mantissa precision
fp8 · 8-bit floating point · Lower memory and higher throughput; quality depends on implementation
int8 · 8-bit integer quantization · Compressed weights/operations with possible quality trade-offs
int4 · 4-bit integer quantization · Aggressive compression; evaluate task quality carefully
unknown · Provider did not publish precision · Undisclosed, not equivalent to full precision

context_tier is standard or extended. Catalog routes with more than 200,000 input tokens are classified as extended unless an explicitly configured route supplies another tier. Always use context_window_tokens for the exact numeric limit.

Benchmark route precision on your real workload before using it as a quality filter.. Use route_id in evaluation logs so precision, provider, region, and model revision stay attributable.. Do not present unknown precision as fp16, bf16, or "full precision".. Use min_context_tokens and preserve_context_window when fallback must not reduce usable context.

Canonical URL: https://mljunction.com/docs/precision-context