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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
impl OwnedQuantizedModel {
/// Generate tokens with zero-allocation inference (IMP-131)
///
/// This is the highest-performance generation path. Uses pre-allocated
/// scratch buffers to eliminate per-token allocations, providing ~3-4x
/// speedup over allocating variants.
///
/// Performance characteristics:
/// - Single allocation at start (scratch buffer + KV cache)
/// - Zero allocations per generated token
/// - ~500KB saved per token for TinyLlama-1.1B
///
/// # Arguments
/// * `prompt` - Input token IDs
/// * `config` - Generation configuration
///
/// # Returns
/// Generated token sequence including prompt
///
/// # Errors
/// Returns error if forward pass fails
pub fn generate_with_scratch(
&self,
prompt: &[u32],
config: &QuantizedGenerateConfig,
) -> Result<Vec<u32>> {
if prompt.is_empty() {
return Err(RealizarError::InvalidShape {
reason: "Prompt cannot be empty".to_string(),
});
}
// GH-167: Check context length before processing to avoid cryptic CUDA errors
if prompt.len() > self.config.context_length {
return Err(RealizarError::ContextLimitExceeded {
provided: prompt.len(),
maximum: self.config.context_length,
});
}
let max_seq_len = prompt.len() + config.max_tokens;
let mut cache = OwnedQuantizedKVCache::from_config(&self.config, max_seq_len);
let mut scratch = InferenceScratchBuffer::from_config(&self.config);
let mut tokens = prompt.to_vec();
// PMAT-TRACE-GGUF-001: Trace config info
if config.trace {
eprintln!(
"[TRACE-CACHE] GGUF scratch: {} layers, hidden_dim={}, vocab={}",
self.config.num_layers, self.config.hidden_dim, self.config.vocab_size
);
eprintln!(
"[TRACE-CACHE] Prefill: {} tokens, max_gen={}",
prompt.len(),
config.max_tokens
);
}
// Process prompt tokens (prefill) - uses scratch buffers
let prefill_start = std::time::Instant::now();
for (pos, &token_id) in prompt.iter().enumerate() {
self.forward_single_with_scratch(token_id, &mut cache, pos, &mut scratch)?;
}
if config.trace {
eprintln!(
"[TRACE-CACHE] Prefill complete: {} tokens in {:?}",
prompt.len(),
prefill_start.elapsed()
);
}
// Generate new tokens - zero allocations per token
// PAR-126: Fixed loop structure to match generate_with_cache:
// 1. Sample from current logits (prefill on first iter, previous forward otherwise)
// 2. Then run forward on the new token to get logits for next iteration
for gen_idx in 0..config.max_tokens {
let token_start = std::time::Instant::now();
// Sample next token from current logits (prefill logits on first iter)
let next_token = if config.temperature == 0.0 || config.top_k == 1 {
ops::argmax(&scratch.logits)
} else {
crate::gguf::OwnedQuantizedModel::sample_topk(
&scratch.logits,
config.temperature,
config.top_k,
)
};
// Check stop condition
if config.stop_tokens.contains(&next_token) {
break;
}
tokens.push(next_token);
// Check max length
if tokens.len() >= max_seq_len {
break;
}
// Get logits for next iteration by forwarding the new token
let position = prompt.len() + gen_idx;
self.forward_single_with_scratch(next_token, &mut cache, position, &mut scratch)?;
// PMAT-TRACE-GGUF-001: Per-token timing
if config.trace {
eprintln!(
"[TRACE-CACHE] pos={}: {} layers took {:?}",
position,
self.config.num_layers,
token_start.elapsed()
);
}
}
Ok(tokens)
}
/// Generate tokens with adaptive CPU/GPU attention (IMP-125)
///
/// This variant of `generate_with_cache` uses `forward_single_with_cache_adaptive`
/// to automatically select between CPU and GPU backends based on cache length.
/// It also records dispatch decisions to the provided metrics tracker.
///
/// # Arguments
/// * `prompt` - Initial token IDs
/// * `config` - Generation configuration
/// * `metrics` - Dispatch metrics tracker for CPU/GPU decision recording
///
/// # Returns
/// Generated token sequence including prompt
///
/// # Errors
/// Returns error if forward pass fails
#[cfg(feature = "gpu")]
pub fn generate_with_cache_adaptive(
&self,
prompt: &[u32],
config: &QuantizedGenerateConfig,
metrics: &std::sync::Arc<DispatchMetrics>,
) -> Result<Vec<u32>> {
if prompt.is_empty() {
return Err(RealizarError::InvalidShape {
reason: "Prompt cannot be empty".to_string(),
});
}
let max_seq_len = prompt.len() + config.max_tokens;
let mut cache = OwnedQuantizedKVCache::from_config(&self.config, max_seq_len);
let mut tokens = prompt.to_vec();
// PMAT-TRACE-GGUF-001: Trace config info
if config.trace {
eprintln!(
"[TRACE-CACHE] GGUF adaptive: {} layers, hidden_dim={}, vocab={}",
self.config.num_layers, self.config.hidden_dim, self.config.vocab_size
);
eprintln!(
"[TRACE-CACHE] Prefill: {} tokens, max_gen={}",
prompt.len(),
config.max_tokens
);
}
// Process prompt tokens (prefill) with adaptive attention
// Keep the logits from the last position for the first generated token
let prefill_start = std::time::Instant::now();
let mut logits = Vec::new();
for (pos, &token_id) in prompt.iter().enumerate() {
logits = self.forward_single_with_cache_adaptive(token_id, &mut cache, pos, metrics)?;
}
if config.trace {
eprintln!(
"[TRACE-CACHE] Prefill complete: {} tokens in {:?}",
prompt.len(),
prefill_start.elapsed()
);
}
// Generate new tokens with adaptive attention
for gen_idx in 0..config.max_tokens {
let token_start = std::time::Instant::now();
// Sample next token from current logits
let next_token = if config.temperature == 0.0 || config.top_k == 1 {
ops::argmax(&logits)
} else {
crate::gguf::OwnedQuantizedModel::sample_topk(
&logits,
config.temperature,
config.top_k,
)
};
// Check stop condition
if config.stop_tokens.contains(&next_token) {
break;
}
tokens.push(next_token);
// Check max length
if tokens.len() >= max_seq_len {
break;
}
// Get logits for next iteration by forwarding the newly sampled token
let position = prompt.len() + gen_idx;
logits =
self.forward_single_with_cache_adaptive(next_token, &mut cache, position, metrics)?;
// PMAT-TRACE-GGUF-001: Per-token timing
if config.trace {
eprintln!(
"[TRACE-CACHE] pos={}: {} layers took {:?}",
position,
self.config.num_layers,
token_start.elapsed()
);
}
}
Ok(tokens)
}
}
include!("fails.rs");