fix: more work on chat and chat titles

This commit is contained in:
2026-05-01 23:34:37 +02:00
parent c495a2ed0a
commit 7db8f6d36b
8 changed files with 296 additions and 22 deletions

View File

@@ -111,7 +111,8 @@ defmodule BDS.AI.Catalog do
def put_model_capabilities(model_id, attrs) when is_binary(model_id) and is_map(attrs) do
capabilities = %{
supports_attachment: truthy?(BDS.MapUtils.attr(attrs, :supports_attachment)),
supports_tool_calls: truthy?(BDS.MapUtils.attr(attrs, :supports_tool_calls))
supports_tool_calls: truthy?(BDS.MapUtils.attr(attrs, :supports_tool_calls)),
disables_reasoning: truthy?(BDS.MapUtils.attr(attrs, :disables_reasoning))
}
put_setting("ai.model_capabilities.#{model_id}", Jason.encode!(capabilities))
@@ -163,7 +164,8 @@ defmodule BDS.AI.Catalog do
@spec model_capabilities(String.t()) :: %{
supports_attachment: boolean(),
supports_tool_calls: boolean()
supports_tool_calls: boolean(),
disables_reasoning: boolean()
}
def model_capabilities(model_id) do
overrides = decode_model_capabilities_override(model_id)
@@ -173,7 +175,8 @@ defmodule BDS.AI.Catalog do
{:ok, model} ->
%{
supports_attachment: model.supports_attachment or "image" in model.input_modalities,
supports_tool_calls: model.supports_tool_calls
supports_tool_calls: model.supports_tool_calls,
disables_reasoning: false
}
_other ->
@@ -196,7 +199,8 @@ defmodule BDS.AI.Catalog do
String.contains?(normalized, "llava"),
supports_tool_calls:
String.contains?(normalized, "gpt") or String.contains?(normalized, "claude") or
String.contains?(normalized, "tool")
String.contains?(normalized, "tool"),
disables_reasoning: false
}
end

View File

@@ -2,6 +2,7 @@ defmodule BDS.AI.Chat do
@moduledoc false
import Ecto.Query
require Logger
alias BDS.AI
alias BDS.AI.Catalog
@@ -23,7 +24,7 @@ defmodule BDS.AI.Chat do
@default_system_prompt "You are the bDS AI backend. Be precise, prefer structured JSON when asked, and avoid inventing blog facts."
@default_max_output_tokens 16_384
@title_max_output_tokens 20
@title_max_output_tokens 256
@chat_title_max_length 30
@chat_max_tool_rounds 10
@default_context_window 128_000
@@ -383,7 +384,7 @@ defmodule BDS.AI.Chat do
conversation = Repo.get!(ChatConversation, conversation_id)
cond do
chat_user_message_count(conversation_id) != 1 ->
chat_user_message_count(conversation_id) < 1 ->
{:ok, reply}
not generated_chat_title?(conversation.title, conversation.model) ->
@@ -418,7 +419,25 @@ defmodule BDS.AI.Chat do
:ok <- Runtime.validate_target(:chat_title, model, mode),
request <- build_chat_title_request(user_content, model),
{:ok, response} <- runtime.generate(Runtime.endpoint_with_model(endpoint, model), request, opts) do
{:ok, sanitize_chat_title(Map.get(response, :content))}
title = sanitize_chat_title(Map.get(response, :content))
if title == "" do
Logger.warning("Chat title generation returned an empty title",
model: model,
content: inspect(Map.get(response, :content)),
usage: inspect(Map.get(response, :usage))
)
end
{:ok, title}
else
{:error, reason} = error ->
Logger.warning("Chat title generation failed", reason: inspect(reason))
error
other ->
Logger.warning("Chat title generation failed", reason: inspect(other))
other
end
end
@@ -431,7 +450,7 @@ defmodule BDS.AI.Chat do
%{
"role" => "system",
"content" =>
"Generate an ultra-short title (2-3 words, max 25 characters) for this conversation. Focus ONLY on the topic. Ignore any capability disclaimers. Output ONLY the title text."
"Generate an ultra-short title (2-3 words, max 25 characters) for this conversation. Focus ONLY on the topic. Ignore any capability disclaimers. Do not include reasoning. Output ONLY the title text."
},
%{"role" => "user", "content" => "Topic: #{String.slice(user_content, 0, 100)}"}
]

View File

@@ -36,6 +36,7 @@ defmodule BDS.AI.OpenAICompatibleRuntime do
"messages" => request.messages,
"max_tokens" => request.max_output_tokens
}
|> maybe_disable_thinking(request.model)
|> maybe_put_tools(Map.get(request, :tools, []))
with {:ok, response} <- HttpClient.post(url, headers, Jason.encode!(payload)),
@@ -136,6 +137,18 @@ defmodule BDS.AI.OpenAICompatibleRuntime do
|> Map.put("tool_choice", "auto")
end
defp maybe_disable_thinking(payload, model) when is_binary(model) do
if BDS.AI.Catalog.model_capabilities(model).disables_reasoning do
Map.update(payload, "chat_template_kwargs", %{"enable_thinking" => false}, fn kwargs ->
Map.put(kwargs || %{}, "enable_thinking", false)
end)
else
payload
end
end
defp maybe_disable_thinking(payload, _model), do: payload
defp normalize_tool_calls(tool_calls) do
Enum.map(tool_calls, fn tool_call ->
%{