Commit
·
6c8acb9
1
Parent(s):
610725e
simplified parallelism system
Browse files- src/index.mts +5 -17
src/index.mts
CHANGED
|
@@ -39,11 +39,9 @@ const maxParallelRequests = 1
|
|
| 39 |
const pending: {
|
| 40 |
total: number;
|
| 41 |
queue: string[];
|
| 42 |
-
aborts: Record<string, any>,
|
| 43 |
} = {
|
| 44 |
total: 0,
|
| 45 |
queue: [],
|
| 46 |
-
aborts: {},
|
| 47 |
}
|
| 48 |
|
| 49 |
const endRequest = (id: string, reason: string) => {
|
|
@@ -51,20 +49,7 @@ const endRequest = (id: string, reason: string) => {
|
|
| 51 |
return
|
| 52 |
}
|
| 53 |
|
| 54 |
-
|
| 55 |
-
try {
|
| 56 |
-
pending.aborts[id].abort()
|
| 57 |
-
} catch (err) {
|
| 58 |
-
console.log(`could not abort request ${id} (${err})`)
|
| 59 |
-
}
|
| 60 |
-
// remove the request from everywhere
|
| 61 |
-
try {
|
| 62 |
-
pending.queue = pending.queue.filter(i => i !== id)
|
| 63 |
-
delete pending.aborts[id]
|
| 64 |
-
console.log(`cleaned up request ${id}`)
|
| 65 |
-
} catch (err) {
|
| 66 |
-
console.log(`failed to properly clean up request ${id}`)
|
| 67 |
-
}
|
| 68 |
console.log(`request ${id} ended (${reason})`)
|
| 69 |
}
|
| 70 |
|
|
@@ -93,7 +78,6 @@ app.get("/", async (req, res) => {
|
|
| 93 |
console.log(`new request ${id}`)
|
| 94 |
|
| 95 |
pending.queue.push(id)
|
| 96 |
-
pending.aborts[id] = new AbortController()
|
| 97 |
|
| 98 |
const prefix = `<html><head>${css}${script}`
|
| 99 |
res.write(prefix)
|
|
@@ -125,7 +109,11 @@ ${prefix}`
|
|
| 125 |
const inputTokens = await llm.tokenize(finalPrompt)
|
| 126 |
console.log("initializing the generator (may take 30s or more)")
|
| 127 |
const generator = await llm.generate(inputTokens)
|
|
|
|
| 128 |
for await (const token of generator) {
|
|
|
|
|
|
|
|
|
|
| 129 |
const tmp = await llm.detokenize(token)
|
| 130 |
process.stdout.write(tmp)
|
| 131 |
res.write(tmp)
|
|
|
|
| 39 |
const pending: {
|
| 40 |
total: number;
|
| 41 |
queue: string[];
|
|
|
|
| 42 |
} = {
|
| 43 |
total: 0,
|
| 44 |
queue: [],
|
|
|
|
| 45 |
}
|
| 46 |
|
| 47 |
const endRequest = (id: string, reason: string) => {
|
|
|
|
| 49 |
return
|
| 50 |
}
|
| 51 |
|
| 52 |
+
pending.queue = pending.queue.filter(i => i !== id)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
console.log(`request ${id} ended (${reason})`)
|
| 54 |
}
|
| 55 |
|
|
|
|
| 78 |
console.log(`new request ${id}`)
|
| 79 |
|
| 80 |
pending.queue.push(id)
|
|
|
|
| 81 |
|
| 82 |
const prefix = `<html><head>${css}${script}`
|
| 83 |
res.write(prefix)
|
|
|
|
| 109 |
const inputTokens = await llm.tokenize(finalPrompt)
|
| 110 |
console.log("initializing the generator (may take 30s or more)")
|
| 111 |
const generator = await llm.generate(inputTokens)
|
| 112 |
+
|
| 113 |
for await (const token of generator) {
|
| 114 |
+
if (!pending.queue.includes(id)) {
|
| 115 |
+
break
|
| 116 |
+
}
|
| 117 |
const tmp = await llm.detokenize(token)
|
| 118 |
process.stdout.write(tmp)
|
| 119 |
res.write(tmp)
|