Instructions to use VladHong/llama-cpp-K2-FAST2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use VladHong/llama-cpp-K2-FAST2 with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: llama cli -hf VladHong/llama-cpp-K2-FAST2
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: llama cli -hf VladHong/llama-cpp-K2-FAST2
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf VladHong/llama-cpp-K2-FAST2 # Run inference directly in the terminal: ./build/bin/llama cli -hf VladHong/llama-cpp-K2-FAST2
Use Docker
docker model run hf.co/VladHong/llama-cpp-K2-FAST2
- Llama new
- LM Studio
- Jan
- Ollama
How to use VladHong/llama-cpp-K2-FAST2 with Ollama:
ollama run hf.co/VladHong/llama-cpp-K2-FAST2
- Unsloth Desktop
- Docker Model Runner
How to use VladHong/llama-cpp-K2-FAST2 with Docker Model Runner:
docker model run hf.co/VladHong/llama-cpp-K2-FAST2
- Lemonade
How to use VladHong/llama-cpp-K2-FAST2 with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull VladHong/llama-cpp-K2-FAST2
Run and chat with the model
lemonade run user.llama-cpp-K2-FAST2-{{QUANT_TAG}}List all available models
lemonade list
- Atomic Chat
File size: 6,028 Bytes
c59fd93 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 | import { ContinueIntentKind, MessageRole, MessageType } from '$lib/enums';
import type { DatabaseMessage } from '$lib/types/database';
import { classifyContinueIntent } from '$lib/utils/agentic';
import { describe, expect, it } from 'vitest';
/**
* Tests for the Continue button intent classifier.
*
* The classifier walks the persisted message history to decide which of three
* resume paths a Continue click should take:
*
* A. append_text -> plain text assistant turn, resume with
* continue_final_message.
* B. rerun_turn -> assistant turn with tool_calls but no tool results yet,
* the stream was cut mid turn and the tool_calls are
* unrecoverable as a token level continuation. Drop the
* target and rerun from the previous history.
* C. next_turn -> assistant turn with tool_calls that were already
* resolved by trailing tool results. Hand the history
* back to the agentic loop so it starts the next turn.
*/
let nextId = 0;
function makeMsg(role: MessageRole, opts: Partial<DatabaseMessage> = {}): DatabaseMessage {
nextId++;
return {
children: [],
content: '',
convId: 'conv-1',
id: `msg-${nextId}`,
parent: null,
role,
timestamp: nextId,
type: MessageType.TEXT,
...opts
};
}
function toolCall(id: string, name: string, args: string = '{}'): string {
return JSON.stringify([{ function: { arguments: args, name }, id, type: 'function' }]);
}
describe('classifyContinueIntent', () => {
it('returns append_text for a plain text assistant turn at the tail', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'hello' }),
makeMsg(MessageRole.ASSISTANT, { content: 'hi there' })
];
const intent = classifyContinueIntent(messages, 1);
expect(intent).toEqual({ kind: ContinueIntentKind.APPEND_TEXT });
});
it('returns append_text for a plain text assistant turn in the middle', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'q1' }),
makeMsg(MessageRole.ASSISTANT, { content: 'a1' }),
makeMsg(MessageRole.USER, { content: 'q2' }),
makeMsg(MessageRole.ASSISTANT, { content: 'a2' })
];
expect(classifyContinueIntent(messages, 1)).toEqual({ kind: ContinueIntentKind.APPEND_TEXT });
});
it('returns rerun_turn when the assistant has tool_calls without results', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'list files' }),
makeMsg(MessageRole.ASSISTANT, {
content: '',
toolCalls: toolCall('call_1', 'bash_tool', '{"command":"ls"}')
})
];
const intent = classifyContinueIntent(messages, 1);
expect(intent).toEqual({ kind: ContinueIntentKind.RERUN_TURN, truncateAfter: 0 });
});
it('returns next_turn when trailing tool results resolve the tool_calls', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'list files' }),
makeMsg(MessageRole.ASSISTANT, {
content: '',
toolCalls: toolCall('call_1', 'bash_tool')
}),
makeMsg(MessageRole.TOOL, { content: 'file1\nfile2', toolCallId: 'call_1' })
];
const intent = classifyContinueIntent(messages, 1);
expect(intent).toEqual({ kind: ContinueIntentKind.NEXT_TURN, truncateAfter: 2 });
});
it('next_turn keeps all consecutive trailing tool results, not just one', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'do many things' }),
makeMsg(MessageRole.ASSISTANT, {
content: '',
toolCalls: JSON.stringify([
{ function: { arguments: '{}', name: 'a' }, id: 'call_1', type: 'function' },
{ function: { arguments: '{}', name: 'b' }, id: 'call_2', type: 'function' }
])
}),
makeMsg(MessageRole.TOOL, { content: 'r1', toolCallId: 'call_1' }),
makeMsg(MessageRole.TOOL, { content: 'r2', toolCallId: 'call_2' })
];
const intent = classifyContinueIntent(messages, 1);
expect(intent).toEqual({ kind: ContinueIntentKind.NEXT_TURN, truncateAfter: 3 });
});
it('next_turn stops at the first non tool message after the target', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'go' }),
makeMsg(MessageRole.ASSISTANT, {
content: '',
toolCalls: toolCall('call_1', 'a')
}),
makeMsg(MessageRole.TOOL, { content: 'r1', toolCallId: 'call_1' }),
makeMsg(MessageRole.USER, { content: 'wait' }),
makeMsg(MessageRole.TOOL, { content: 'late', toolCallId: 'call_1' })
];
const intent = classifyContinueIntent(messages, 1);
// truncateAfter must point at the contiguous tool block, not jump over
// the user message to grab the dangling late tool.
expect(intent).toEqual({ kind: ContinueIntentKind.NEXT_TURN, truncateAfter: 2 });
});
it('returns append_text when toolCalls is set but parses to empty array', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'q' }),
makeMsg(MessageRole.ASSISTANT, { content: 'a', toolCalls: '[]' })
];
expect(classifyContinueIntent(messages, 1)).toEqual({ kind: ContinueIntentKind.APPEND_TEXT });
});
it('returns append_text when toolCalls is malformed JSON', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'q' }),
makeMsg(MessageRole.ASSISTANT, { content: 'a', toolCalls: '{not json' })
];
expect(classifyContinueIntent(messages, 1)).toEqual({ kind: ContinueIntentKind.APPEND_TEXT });
});
it('returns append_text defensively when idx points at a non assistant message', () => {
const messages = [
makeMsg(MessageRole.USER, { content: 'q' }),
makeMsg(MessageRole.ASSISTANT, { content: 'a' })
];
expect(classifyContinueIntent(messages, 0)).toEqual({ kind: ContinueIntentKind.APPEND_TEXT });
});
it('returns append_text defensively when idx is out of bounds', () => {
const messages = [makeMsg(MessageRole.ASSISTANT, { content: 'a' })];
expect(classifyContinueIntent(messages, 5)).toEqual({ kind: ContinueIntentKind.APPEND_TEXT });
expect(classifyContinueIntent([], 0)).toEqual({ kind: ContinueIntentKind.APPEND_TEXT });
});
});
|