Files
wbin90620 34b72da2c5 model, object: no panic on a local upstream or a completion model, each token priced once
hanzo-inc/cloud#460 (moved from #116) listed eight defects at c7f85eed. Seven
were still on main; this fixes them.

A LocalModelProvider of type Zen, Enso or Hanzo had no client, and neither did
DeepSeek with a model other than chat or reasoner, so the first call
dereferenced nil. dialect() now names each type's client and flush once, with
typed values in place of an `any` asserted in each loop, and QueryText refuses
a type with no client. That also ends the assertion panics: OpenAI completion
models asserted the answer-only flush while holding the think one, and so did
Local and DigitalOcean completion models in local.go. Both completion loops
now skip a chunk with no choice, and local.go's no longer returns a nil result
with no error when the count fails.

DeepSeek, Mistral, SiliconFlow, Baichuan and BaiduCloud priced the total token
count at both the input and the output rate. The prompt is now priced at the
input rate and the answer at the output rate. gpt-4.5's case sat after
gpt-4's, which also matches it, so gpt-4.5 billed as gpt-4; it now comes first
and bills at the rate the pricing table publishes.

GetMessagesForChats bound the chat names to a bare "{:p0}" WHERE; it is
dbx.In("chat", ...). updateErrorText's branches were inverted, so a record
with no id updated WHERE id = 0. The Guacamole handshake returned nil on a
failed write, and UploadVideo returned "", nil on each failure; both return
the error.

Item 7 (gateway frames) is left: the gateway's own errors are gateway frames
now, but its chat, models and balance fast paths still answer with cloud
frames. That needs hanzoai/gateway's side confirmed first.
2026-11-09 04:15:19 -05:00

116 lines
4.7 KiB
Go

// Copyright 2023-2025 Hanzo AI Inc. All Rights Reserved.
// Portions Copyright 2025 The OpenAgent Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package model
import (
"fmt"
"io"
"strings"
"github.com/gage-technologies/mistral-go"
"github.com/hanzoai/ai/i18n"
)
type MistralModelProvider struct {
client *mistral.MistralClient
modelName string
}
func NewMistralProvider(apiKey, modelName string) (*MistralModelProvider, error) {
client := mistral.NewMistralClientDefault(apiKey)
return &MistralModelProvider{
client: client,
modelName: modelName,
}, nil
}
func (c *MistralModelProvider) GetPricing() string {
return `URL: https://mistral.ai/technology/#pricing
| Model | Input Price($) per 1K tokens | Output Price($) per 1K tokens |
|-------------------------------------|-------------------------------|--------------------------------|
| mistral-large-latest | 0.002 | 0.006 |
| pixtral-large-latest | 0.002 | 0.006 |
| mistral-small-latest | 0.0002 | 0.0006 |
| codestral-latest | 0.0003 | 0.0009 |
| ministral-8b-latest | 0.0001 | 0.0001 |
| ministral-3b-latest | 0.00004 | 0.00004 |
| pixtral-12b | 0.00015 | 0.00015 |
| mistral-nemo | 0.00015 | 0.00015 |
| open-mistral-7b | 0.00025 | 0.00025 |
| open-mixtral-8x7b | 0.0007 | 0.0007 |
| open-mixtral-8x22b | 0.002 | 0.006 |
`
}
func (c *MistralModelProvider) calculatePrice(modelResult *ModelResult, lang string) error {
price := 0.0
priceTable := map[string][2]float64{
"mistral-large-latest": {0.002, 0.006},
"pixtral-large-latest": {0.002, 0.006},
"mistral-small-latest": {0.0002, 0.0006},
"codestral-latest": {0.0003, 0.0009},
"ministral-8b-latest": {0.0001, 0.0001},
"ministral-3b-latest": {0.00004, 0.0001},
"pixtral-12b": {0.00015, 0.00015},
"mistral-nemo": {0.00015, 0.00015},
"open-mistral-7b": {0.00025, 0.00025},
"open-mixtral-8x7b ": {0.002, 0.002},
"open-mixtral-8x22b": {0.002, 0.006},
}
if priceItem, ok := priceTable[c.modelName]; ok {
inputPrice := getPrice(modelResult.PromptTokenCount, priceItem[0])
outputPrice := getPrice(modelResult.ResponseTokenCount, priceItem[1])
price = inputPrice + outputPrice
} else {
return fmt.Errorf("%s", fmt.Sprintf(i18n.Translate(lang, "embedding:calculatePrice() error: unknown model type: %s"), c.modelName))
}
modelResult.TotalPrice = price
modelResult.Currency = "USD"
return nil
}
func (c *MistralModelProvider) QueryText(question string, writer io.Writer, history []*RawMessage, prompt string, knowledgeMessages []*RawMessage, agentInfo *AgentInfo, lang string) (*ModelResult, error) {
chatRes, err := c.client.Chat(c.modelName, []mistral.ChatMessage{{Content: question, Role: mistral.RoleUser}}, nil)
if err != nil {
return nil, fmt.Errorf("%s", fmt.Sprintf(i18n.Translate(lang, "model:error getting chat completion: %v"), err))
}
respText := chatRes.Choices[0].Message.Content
respText = strings.TrimSpace(respText)
_, err = fmt.Fprint(writer, respText)
if err != nil {
return nil, fmt.Errorf("%s", fmt.Sprintf(i18n.Translate(lang, "model:failed to write response: %v"), err))
}
modelResult, err := getDefaultModelResult(c.modelName, question, respText)
if err != nil {
return nil, err
}
err = c.calculatePrice(modelResult, lang)
if err != nil {
return nil, fmt.Errorf("%s", fmt.Sprintf(i18n.Translate(lang, "embedding:failed to calculate price: %v"), err))
}
modelResult.PromptTokenCount += len(question)
return modelResult, nil
}