chore: expurgo de blobs excedentes e realinhamento de estado
release.yml (push) Invalid workflow file

This commit is contained in:
Evandro Fonseca Junior committed 2026-02-22 14:06:01 -03:00
1 parent 593efd42e0
commit a7cbd1a3b4
114 files changed
+20716 -4258

No files matched your search

+245 -5
View File
@@ -1,10 +1,10 @@
use serde::{Deserialize, Serialize};
/// Mensagem para o Ollama API
#[derive(Debug, Serialize)]
struct OllamaMessage {
role: String,
content: String,
#[derive(Debug, Serialize, Deserialize, Clone)]
pub struct OllamaMessage {
pub role: String,
pub content: String,
}
/// Request para chat do Ollama
@@ -70,6 +70,9 @@ impl OllamaClient {
system_prompt: Option<&str>,
user_prompt: &str,
) -> Result<String, String> {
log::info!("[LLM] Headless query model={} system_prompt_present={} user_prompt_len={}", model, system_prompt.is_some(), user_prompt.len());
if let Some(sys) = system_prompt { log::info!("[LLM] System Prompt:\n{}", sys); }
log::info!("[LLM] User Prompt:\n{}", user_prompt);
// Verificar conexão primeiro
self.check_connection().await?;
@@ -96,12 +99,15 @@ impl OllamaClient {
};
let url = format!("{}/api/chat", self.base_url);
log::info!("[LLM_DEBUG] Iniciando request HTTP para o modelo (chat stream)...");
let response = self.client
.post(&url)
.json(&request)
.timeout(std::time::Duration::from_secs(30))
.send()
.await
.map_err(|e| format!("Failed to send request to Ollama: {}", e))?;
log::info!("[LLM] HTTP response received (chat stream). Status: {}", response.status());
if !response.status().is_success() {
return Err(format!("Ollama returned status: {}", response.status()));
@@ -113,9 +119,14 @@ impl OllamaClient {
use futures_util::StreamExt;
let mut logged_first_chunk = false;
while let Some(chunk) = stream.next().await {
let chunk = chunk.map_err(|e| format!("Failed to read chunk: {}", e))?;
let text = String::from_utf8_lossy(&chunk);
if !logged_first_chunk {
log::debug!("[LLM] First chunk received ({} bytes)", text.len());
logged_first_chunk = true;
}
// Processar cada linha (Ollama envia JSON por linha)
for line in text.lines() {
@@ -141,7 +152,7 @@ impl OllamaClient {
}
}
}
log::info!("[LLM] Headless Raw Output ({} chars):\n{}", full_response.len(), full_response.trim());
if full_response.is_empty() {
return Err("Empty response from Ollama".to_string());
}
@@ -252,5 +263,234 @@ impl OllamaClient {
"💬".to_string() // Default
}
}
/// Gera completion usando /api/generate (não streaming)
pub async fn generate_completion(
&self,
model: &str,
prompt: &str,
options: Option<GenerationOptions>,
) -> Result<String, String> {
log::info!("[LLM] generate_completion model={} prompt_len={}", model, prompt.len());
log::info!("[LLM] Prompt:\n{}", prompt);
self.check_connection().await?;
let mut request_body = serde_json::json!({
"model": model,
"prompt": prompt,
"stream": false,
});
// Adicionar options se fornecidas
if let Some(opts) = options.as_ref() {
let mut opts_json = serde_json::Map::new();
if let Some(temp) = opts.temperature {
// serde_json::Number não implementa From<f64>, usar json! macro ou Number::from_f64
if let Some(num) = serde_json::Number::from_f64(temp) {
opts_json.insert("temperature".to_string(), serde_json::Value::Number(num));
} else {
// Fallback: usar como string se não conseguir converter
opts_json.insert("temperature".to_string(), serde_json::Value::String(temp.to_string()));
}
}
if let Some(num_predict) = opts.num_predict {
opts_json.insert("num_predict".to_string(), serde_json::Value::Number(num_predict.into()));
}
if let Some(format) = opts.format.clone() {
opts_json.insert("format".to_string(), serde_json::Value::String(format));
}
if !opts_json.is_empty() {
request_body["options"] = serde_json::Value::Object(opts_json);
}
log::info!("[LLM] Options: temperature={:?} num_predict={:?} format={:?}", opts.temperature, opts.num_predict, opts.format.as_deref());
}
let url = format!("{}/api/generate", self.base_url);
log::info!("[LLM_DEBUG] Iniciando request HTTP para o modelo (generate)...");
let response = self.client
.post(&url)
.json(&request_body)
.send()
.await
.map_err(|e| format!("Failed to send request to Ollama: {}", e))?;
log::info!("[LLM] HTTP response received (generate). Status: {}", response.status());
if !response.status().is_success() {
return Err(format!("Ollama returned status: {}", response.status()));
}
#[derive(serde::Deserialize)]
struct GenerateResponse {
response: String,
}
let body_text = response.text()
.await
.map_err(|e| format!("Failed to read response text: {}", e))?;
log::info!("[LLM] Raw HTTP Body ({} chars)", body_text.len());
let result: GenerateResponse = match serde_json::from_str::<GenerateResponse>(&body_text) {
Ok(val) => val,
Err(e) => {
log::error!("[LLM] Failed to parse JSON body: {}", e);
log::error!("[LLM] Body content: {}", body_text);
return Err(format!("Failed to parse response: {}", e));
}
};
let mut raw = result.response.trim().to_string();
if let Ok(re) = regex::Regex::new(r"(?s)<think>.*?</think>") { raw = re.replace_all(&raw, "").to_string(); }
if let Ok(re) = regex::Regex::new(r"(?i)^\s*(here\s+is\s+the\s+query|sure|okay|final\s+query)\s*[:,-]*\s*") { raw = re.replace(&raw, "").to_string(); }
if let Ok(re) = regex::Regex::new(r"^\s*```[a-zA-Z]*\s*") { raw = re.replace(&raw, "").to_string(); }
if let Ok(re) = regex::Regex::new(r"\s*```\s*$") { raw = re.replace(&raw, "").to_string(); }
raw = raw.trim().to_string();
log::info!("[LLM] Raw Output ({} chars):\n{}", raw.len(), raw);
if let Some(fmt) = options.as_ref().and_then(|o| o.format.as_ref()) {
if fmt.to_lowercase() == "json" {
match serde_json::from_str::<serde_json::Value>(&raw) {
Ok(val) => {
let kind = if val.is_array() { "array" } else if val.is_object() { "object" } else { "other" };
log::info!("[LLM] JSON parse ok ({})", kind);
}
Err(e) => {
log::warn!("[LLM] JSON parse failed: {}", e);
}
}
}
}
Ok(raw)
}
/// Gera chat completion usando /api/chat (não streaming)
pub async fn generate_chat_completion(
&self,
model: &str,
messages: Vec<OllamaMessage>,
options: Option<GenerationOptions>,
) -> Result<String, String> {
let msgs_json = serde_json::to_string(&messages).unwrap_or_else(|_| "[]".to_string());
log::info!("[LLM] generate_chat_completion model={} messages_len={} messages_json={} options_temp={:?} options_num_predict={:?}", model, messages.len(), msgs_json, options.as_ref().and_then(|o| o.temperature), options.as_ref().and_then(|o| o.num_predict));
self.check_connection().await?;
let mut request_body = serde_json::json!({
"model": model,
"messages": messages,
"stream": false,
});
// Adicionar options se fornecidas
if let Some(opts) = options {
let mut opts_json = serde_json::Map::new();
if let Some(temp) = opts.temperature {
// serde_json::Number não implementa From<f64>, usar Number::from_f64
if let Some(num) = serde_json::Number::from_f64(temp) {
opts_json.insert("temperature".to_string(), serde_json::Value::Number(num));
} else {
// Fallback: usar como string se não conseguir converter
opts_json.insert("temperature".to_string(), serde_json::Value::String(temp.to_string()));
}
}
if let Some(num_predict) = opts.num_predict {
opts_json.insert("num_predict".to_string(), serde_json::Value::Number(num_predict.into()));
}
if !opts_json.is_empty() {
request_body["options"] = serde_json::Value::Object(opts_json);
}
}
let url = format!("{}/api/chat", self.base_url);
let response = self.client
.post(&url)
.json(&request_body)
.send()
.await
.map_err(|e| format!("Failed to send request to Ollama: {}", e))?;
if !response.status().is_success() {
return Err(format!("Ollama returned status: {}", response.status()));
}
#[derive(serde::Deserialize)]
struct ChatResponse {
message: OllamaMessageResponse,
}
let result: ChatResponse = response.json()
.await
.map_err(|e| format!("Failed to parse response: {}", e))?;
let raw = result.message.content.trim().to_string();
log::info!("[LLM] Chat Raw Output ({} chars):\n{}", raw.len(), raw);
Ok(raw)
}
/// Obtém informações de um modelo usando /api/show
pub async fn get_model_info(&self, model_name: &str) -> Result<serde_json::Value, String> {
self.check_connection().await?;
let url = format!("{}/api/show", self.base_url);
let response = self.client
.post(&url)
.json(&serde_json::json!({ "name": model_name }))
.send()
.await
.map_err(|e| format!("Failed to connect to Ollama: {}", e))?;
if !response.status().is_success() {
return Err(format!("Ollama returned status: {}", response.status()));
}
let result: serde_json::Value = response.json()
.await
.map_err(|e| format!("Failed to parse response: {}", e))?;
Ok(result)
}
/// Lista modelos disponíveis usando /api/tags
pub async fn get_tags(&self) -> Result<Vec<OllamaModel>, String> {
self.check_connection().await?;
let url = format!("{}/api/tags", self.base_url);
let response = self.client
.get(&url)
.send()
.await
.map_err(|e| format!("Failed to connect to Ollama: {}", e))?;
if !response.status().is_success() {
return Err(format!("Ollama returned status: {}", response.status()));
}
#[derive(serde::Deserialize)]
struct TagsResponse {
models: Vec<OllamaModel>,
}
let result: TagsResponse = response.json()
.await
.map_err(|e| format!("Failed to parse response: {}", e))?;
Ok(result.models)
}
}
/// Opções para geração de texto
#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct GenerationOptions {
#[serde(skip_serializing_if = "Option::is_none")]
pub temperature: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub num_predict: Option<u32>,
#[serde(skip_serializing_if = "Option::is_none")]
pub format: Option<String>,
}
/// Modelo Ollama retornado por /api/tags
#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct OllamaModel {
pub name: String,
#[serde(default)]
pub size: Option<u64>,
#[serde(default)]
pub modified_at: Option<String>,
#[serde(default)]
pub digest: Option<String>,
}