Evan
6785807633
Refs librefang/librefang#4842 — long-term replacement for the substring match that the OpenAI driver currently uses to decide how to handle `reasoning_content` on historical assistant turns. Three provider-specific behaviours that the driver must distinguish at wire time, now expressed as catalog metadata: * `strip` — DeepSeek R1 / deepseek-reasoner. The API rejects requests that carry reasoning_content on previous assistant messages. * `echo` — DeepSeek V4 Flash. Thinking mode is on by default and the API rejects multi-turn requests when assistant turns containing tool_calls don't echo back the original reasoning text. This is the bug surfaced in librefang/librefang#4842. * `empty_string` — Moonshot / Kimi K2 family. The field must be present (empty string) on tool_calls turns, with thinking disabled wire-side for multi-turn compatibility. * `none` (default) — most providers; field is omitted entirely. V4 Pro is intentionally NOT marked `echo` — librefang#4842 reports it working out-of-the-box; flip when there's an empirical reproducer. Marks affected models: providers/deepseek.toml deepseek-v4-flash → echo deepseek-reasoner → strip providers/moonshot.toml kimi-k2.6, kimi-k2.5, kimi-k2 → empty_string providers/kimi-coding.toml kimi-for-coding → empty_string providers/byteplus-coding.toml kimi-k2.5 → empty_string providers/novita.toml moonshotai/kimi-k2-thinking → empty_string Tooling: * schema.toml registers the field with the four enum options and a `none` default so existing TOML files keep parsing unchanged. * scripts/validate.py rejects unknown enum values; verified with a hand-crafted negative case (`reasoning_echo_policy = "bogus"` → validation fails with the expected message). * `python3 scripts/validate.py` passes (267 models). The librefang side that consumes this field will land in a follow-up PR — until then, registry consumers ignore the field via `#[serde(default)]` and the existing substring fallback continues to work, so this commit is safe to ship independently.
72 lines
1.8 KiB
TOML
72 lines
1.8 KiB
TOML
# DeepSeek — https://deepseek.com
|
|
# Models: 4
|
|
|
|
[provider]
|
|
id = "deepseek"
|
|
display_name = "DeepSeek"
|
|
api_key_env = "DEEPSEEK_API_KEY"
|
|
base_url = "https://api.deepseek.com/v1"
|
|
key_required = true
|
|
|
|
[[models]]
|
|
id = "deepseek-v4-pro"
|
|
display_name = "DeepSeek V4 Pro"
|
|
tier = "frontier"
|
|
context_window = 1000000
|
|
max_output_tokens = 384000
|
|
input_cost_per_m = 0.435
|
|
output_cost_per_m = 0.87
|
|
supports_tools = true
|
|
supports_vision = false
|
|
supports_streaming = true
|
|
supports_thinking = true
|
|
aliases = ["deepseek-v4", "deepseek-pro"]
|
|
|
|
[[models]]
|
|
id = "deepseek-v4-flash"
|
|
display_name = "DeepSeek V4 Flash"
|
|
tier = "smart"
|
|
context_window = 1000000
|
|
max_output_tokens = 384000
|
|
input_cost_per_m = 0.14
|
|
output_cost_per_m = 0.28
|
|
supports_tools = true
|
|
supports_vision = false
|
|
supports_streaming = true
|
|
supports_thinking = true
|
|
# Thinking mode is on by default and the API rejects multi-turn requests
|
|
# where assistant turns containing tool_calls don't echo back the original
|
|
# reasoning_content. See librefang/librefang#4842.
|
|
reasoning_echo_policy = "echo"
|
|
aliases = ["deepseek-flash"]
|
|
|
|
[[models]]
|
|
id = "deepseek-chat"
|
|
display_name = "DeepSeek V3"
|
|
tier = "smart"
|
|
context_window = 64000
|
|
max_output_tokens = 8192
|
|
input_cost_per_m = 0.32
|
|
output_cost_per_m = 0.89
|
|
supports_tools = true
|
|
supports_vision = false
|
|
supports_streaming = true
|
|
aliases = ["deepseek", "deepseek-v3"]
|
|
|
|
[[models]]
|
|
id = "deepseek-reasoner"
|
|
display_name = "DeepSeek R1"
|
|
tier = "frontier"
|
|
context_window = 64000
|
|
max_output_tokens = 8192
|
|
input_cost_per_m = 0.55
|
|
output_cost_per_m = 2.19
|
|
supports_tools = false
|
|
supports_vision = false
|
|
supports_streaming = true
|
|
supports_thinking = true
|
|
# R1 returns reasoning_content in responses but the API rejects multi-turn
|
|
# requests that carry it on previous assistant messages — drivers must strip.
|
|
reasoning_echo_policy = "strip"
|
|
aliases = ["deepseek-r1"]
|