Files
deepseek-harness/docs/core-data-structures/llm-streaming.md
Tianyi Cui 2745879132 refactor(llm): drop the image content block until a path can honor it
ImageBlock had no production producer and every consumer dropped it:
the deepseek serializer skipped it, the pi-ai converter skipped it as
unrepresentable, the ACP bridge neither advertises image prompt
capability nor forwards image blocks, and compact-basic charged a flat
85-token estimate and rendered an [image] placeholder. A block
constructed today would silently vanish from the wire — the vocabulary
advertised a capability no path honors, the silent-data-loss shape the
defensive patterns warn against. The only constructors were tests
pinning the skip/estimate branches.

Remove ImageBlock and its ContentBlockMap entry (its cache?: CacheHint
field leaves with it; CacheHint itself and the other two cache? fields
are out of scope). compact-basic loses its explicit image estimate and
placeholder arms (the merge-extensible default arms absorb the case);
the deepseek serializer, pi-ai converter, and ACP codec already handled
image in their default arms, so only their image-naming comments
change. The codec's inbound rejection of ACP-protocol image prompt
content stays — that guards wire content a client can send regardless
of our vocabulary.

Tests that constructed harness image blocks to pin the removed branches
are dropped (the 85-token estimate pin) or retargeted onto plugin-added
block types / other non-text blocks, which the surviving default arms
own. Docs, the type-equiv pastes, and the content-block vocabulary
RFC's block list and multimodal-home consequence are updated in the
same change; the RFC moves to implemented/ and the index is
regenerated. A real multimodal feature reintroduces image via
declaration merging together with the adapter mapping, ACP
advertisement, and compaction pricing that honor it.
2026-07-04 17:21:13 +08:00

66 lines
4.3 KiB
Markdown

# LLM Streaming
The wire-level streaming vocabulary of [dsh-llm](../../packages/llm/llm). [core.md](core.md) introduces `StreamChunk`, `Message`, and `ContentBlock`; this page owns the full chunk protocol, the adapter contract every adapter must obey, and the shared assembler.
Source: [`packages/llm/llm/src/types.ts`](../../packages/llm/llm/src/types.ts)
## `StreamChunk` — the raw protocol
A streaming response interleaves several typed blocks (text, reasoning, multiple tool calls). `index` ties each delta to its block; `block-end` carries the fully-assembled `ContentBlock` so consumers don't have to re-assemble deltas themselves. It is a **closed** discriminated union — a `switch` over `type` ends with `assertNever`, so adding a variant breaks compilation at every consumer that must handle it.
```ts type-equiv
type StreamChunk =
| { type: 'block-start'; index: number; blockType: ContentBlockType }
| { type: 'text-delta'; index: number; text: string }
| { type: 'reasoning-delta'; index: number; text: string }
| { type: 'tool-call-delta'; index: number; id: CallId; name?: string; argumentsDelta: string }
| { type: 'block-end'; index: number; block: ContentBlock }
| { type: 'usage'; usage: TokenUsage }
| { type: 'finish'; reason: FinishReason }
```
## The adapter contract
Every adapter MUST obey these, and every consumer may rely on them:
- **`usage` before `finish`, nothing after `finish`.** Defer both to the provider's end-of-stream marker so a trailing usage-only chunk can't violate the ordering.
- **Tool-call `arguments` stay raw JSON strings end-to-end.** Partial fragments stream via `argumentsDelta`; a provider that hands back parsed objects re-stringifies at `block-end`.
- **Two sanctioned error paths.** A failure may either THROW from `stream()` (transport/protocol errors) **or** end the stream with `finish {kind:'error'|'aborted'}` (provider in-band errors, for adapters that can't throw mid-stream). Consumers must handle *both*. The agent loop translates a finish-error/aborted into a turn error — it never logs a normal completed assistant message for a failed step.
This contract is why two adapters exist as a deliberate pair: `dsh-llm-deepseek` (hand-rolled fetch/SSE) and `dsh-llm-pi-ai` (the same endpoint through `@earendil-works/pi-ai`). Two independent internals over one contract is what pinned the protocol down — the library-backed adapter can't throw mid-stream, so it exercises the finish-chunk error path the hand-rolled one might not.
## `TokenUsage`
Per-call token accounting. Counts are **disjoint**: `inputTokens` is uncached input only; cached input is reported separately, and billed input is the sum of the three. Adapters whose providers fold cache hits into a single prompt total (DeepSeek's `prompt_tokens`) subtract them back out.
```ts type-equiv
interface TokenUsage {
inputTokens: number
outputTokens: number
cacheReadTokens?: number
cacheWriteTokens?: number
reasoningTokens?: number
}
```
## `BlockAssembler`
`BlockAssembler` ([`packages/llm/llm/src/assembler.ts`](../../packages/llm/llm/src/assembler.ts)) is the single shared implementation that folds a `StreamChunk` stream back into `ContentBlock`s and a final `Message`. The loop logs the raw chunks (for replay fidelity) while feeding the same chunks through an assembler — so the canonical log keeps token-level detail and the derived message is rebuilt deterministically. A consumer that needs the assembled result without re-implementing the fold uses this.
## The seam
`LlmAdapter` is the provider seam: subclass, implement `stream()`, register with `ctx.llm.registerAdapter(models, adapter)`. The `block-start` / `block-end` `index` correlation and the assembler together mean an adapter only has to emit well-formed chunks — block reassembly is not each adapter's problem. The consumer surface (`ctx.llm.stream()`) and the `llm/stream` waterfall are described in [architecture.md § The vocabulary](../architecture.md#the-vocabulary-dsh-llm).
`ContentBlockType` (the key set the `index`-correlated blocks carry) derives from `ContentBlockMap`:
```ts type-equiv
interface ContentBlockMap {
'text': TextBlock
'reasoning': ReasoningBlock
'tool-call': ToolCallBlock
'tool-result': ToolResultBlock
}
```
See [core.md § Content blocks and messages](core.md#content-blocks-and-messages) for the block interfaces.