Instructions to use dappalumbo91/fluid-language-model with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use dappalumbo91/fluid-language-model with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="dappalumbo91/fluid-language-model", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("dappalumbo91/fluid-language-model", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use dappalumbo91/fluid-language-model with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "dappalumbo91/fluid-language-model" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "dappalumbo91/fluid-language-model", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/dappalumbo91/fluid-language-model
- SGLang
How to use dappalumbo91/fluid-language-model with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "dappalumbo91/fluid-language-model" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "dappalumbo91/fluid-language-model", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "dappalumbo91/fluid-language-model" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "dappalumbo91/fluid-language-model", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use dappalumbo91/fluid-language-model with Docker Model Runner:
docker model run hf.co/dappalumbo91/fluid-language-model
Powers from the written product
Browse filesSquares 44/44. Higher exponents 45/45. Divide-back 44/44. 2^13 is 8192. 12^3 is 1728. 37^2 is 1369.
README.md
CHANGED
|
@@ -43,6 +43,12 @@ Far products were 225/225. The written product plus 1 was 225/225. The written p
|
|
| 43 |
|
| 44 |
`37*37` wrote `42+7=49`, then place `0+0` as 49. The partial sums were `0+49=49`, `49+210=259`, `259+210=469`, and `469+900=1369`. The `49` on the next line is the token list `[2, 20, 15, 11, 11]` that the previous line wrote. The written 1369 plus 1 is 1370. The written 1369 divided by 37 is 37.
|
| 45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
## Load
|
| 47 |
|
| 48 |
```python
|
|
|
|
| 43 |
|
| 44 |
`37*37` wrote `42+7=49`, then place `0+0` as 49. The partial sums were `0+49=49`, `49+210=259`, `259+210=469`, and `469+900=1369`. The `49` on the next line is the token list `[2, 20, 15, 11, 11]` that the previous line wrote. The written 1369 plus 1 is 1370. The written 1369 divided by 37 is 37.
|
| 45 |
|
| 46 |
+
## Powers
|
| 47 |
+
|
| 48 |
+
`python -u fluid.py power` on an NVIDIA GeForce RTX 5070, 2026-11-10. A power multiplies the written product by the base again. Bases 2 through 45. A product that would not fit in four places is not attempted. The last written power is divided by the base, and that quotient is the previous line.
|
| 49 |
+
|
| 50 |
+
Squares were 44/44. Higher exponents were 45/45. Divide-back was 44/44. `37^2` is 1369, and `1369/37` is 37. `12^2` is 144. `12^3` wrote `728+1000=1728`, and `1728/12` is 144. `2^13` is 8192, and `8192/2` is 4096.
|
| 51 |
+
|
| 52 |
## Load
|
| 53 |
|
| 54 |
```python
|
fluid.py
CHANGED
|
@@ -23,8 +23,14 @@ wrote. A place shift is applied to those tokens, and the shifted tokens are
|
|
| 23 |
a later right operand. The command scores far products. It does not train
|
| 24 |
and it does not write a checkpoint.
|
| 25 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
python -u fluid.py
|
| 27 |
python -u fluid.py think
|
|
|
|
| 28 |
"""
|
| 29 |
|
| 30 |
from __future__ import annotations
|
|
@@ -408,6 +414,143 @@ def _report_thought() -> None:
|
|
| 408 |
print(f"thought product/right {div_hits}/{len(rows)}", flush=True)
|
| 409 |
|
| 410 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 411 |
def _chain(model: FluidForCausalLM, rows: list[tuple[int, int, int, int]], extra, device: torch.device) -> tuple[int, int]:
|
| 412 |
prompts = _prompts(rows, device)
|
| 413 |
first = _generate_answer(model, prompts)
|
|
@@ -537,5 +680,7 @@ def main() -> None:
|
|
| 537 |
if __name__ == "__main__":
|
| 538 |
if len(sys.argv) > 1 and sys.argv[1] == "think":
|
| 539 |
_report_thought()
|
|
|
|
|
|
|
| 540 |
else:
|
| 541 |
main()
|
|
|
|
| 23 |
a later right operand. The command scores far products. It does not train
|
| 24 |
and it does not write a checkpoint.
|
| 25 |
|
| 26 |
+
``power`` multiplies that written product by the base again. A result that
|
| 27 |
+
would not fit in four places is not attempted. The last written power is
|
| 28 |
+
then divided by the base, and that quotient has to be the previous line.
|
| 29 |
+
The command does not train and it does not write a checkpoint.
|
| 30 |
+
|
| 31 |
python -u fluid.py
|
| 32 |
python -u fluid.py think
|
| 33 |
+
python -u fluid.py power
|
| 34 |
"""
|
| 35 |
|
| 36 |
from __future__ import annotations
|
|
|
|
| 414 |
print(f"thought product/right {div_hits}/{len(rows)}", flush=True)
|
| 415 |
|
| 416 |
|
| 417 |
+
def _multiply_written(
|
| 418 |
+
model: FluidForCausalLM,
|
| 419 |
+
lefts: list[list[int]],
|
| 420 |
+
rights: list[list[int]],
|
| 421 |
+
device: torch.device,
|
| 422 |
+
probe: set[tuple[int, int]] | None = None,
|
| 423 |
+
) -> list[list[int]]:
|
| 424 |
+
"""Multiply written answers. The next addition reads the tokens just written."""
|
| 425 |
+
count = len(lefts)
|
| 426 |
+
left_digits = [place._digits_of(tokens) for tokens in lefts]
|
| 427 |
+
right_digits = [place._digits_of(tokens) for tokens in rights]
|
| 428 |
+
partials: list[list[list[int]]] = [[] for _ in range(count)]
|
| 429 |
+
watched = probe or set()
|
| 430 |
+
for i in range(place.PLACES):
|
| 431 |
+
for j in range(place.PLACES):
|
| 432 |
+
active = [
|
| 433 |
+
row
|
| 434 |
+
for row in range(count)
|
| 435 |
+
if left_digits[row][i] != 0 and right_digits[row][j] != 0
|
| 436 |
+
]
|
| 437 |
+
if not active:
|
| 438 |
+
continue
|
| 439 |
+
print(f"power mul places {i}+{j} rows {len(active)}", flush=True)
|
| 440 |
+
totals = [scratch._number(0) for _ in active]
|
| 441 |
+
addends = [scratch._number(left_digits[row][i]) for row in active]
|
| 442 |
+
remaining = [right_digits[row][j] for row in active]
|
| 443 |
+
while any(item > 0 for item in remaining):
|
| 444 |
+
todo = [index for index, item in enumerate(remaining) if item > 0]
|
| 445 |
+
before = [totals[index] for index in todo]
|
| 446 |
+
step_rights = [addends[index] for index in todo]
|
| 447 |
+
wrote = _write_lines(model, before, step_rights, scratch.OP_ADD, device)
|
| 448 |
+
for index, answer, left_tokens, right_tokens in zip(todo, wrote, before, step_rights):
|
| 449 |
+
pair = (scratch._read_number(lefts[active[index]]), scratch._read_number(rights[active[index]]))
|
| 450 |
+
if pair in watched:
|
| 451 |
+
print(
|
| 452 |
+
f" {scratch._read_number(left_tokens)}+{scratch._read_number(right_tokens)}={scratch._read_number(answer)}",
|
| 453 |
+
flush=True,
|
| 454 |
+
)
|
| 455 |
+
totals[index] = answer
|
| 456 |
+
remaining[index] -= 1
|
| 457 |
+
for index, row in enumerate(active):
|
| 458 |
+
shifted = place._answer(place._shift_places(place._digits_of(totals[index]), i + j))
|
| 459 |
+
partials[row].append(shifted)
|
| 460 |
+
pair = (scratch._read_number(lefts[row]), scratch._read_number(rights[row]))
|
| 461 |
+
if pair in watched:
|
| 462 |
+
print(
|
| 463 |
+
f" place {i}+{j} -> {scratch._read_number(shifted)} tokens {shifted}",
|
| 464 |
+
flush=True,
|
| 465 |
+
)
|
| 466 |
+
running = [scratch._number(0) for _ in range(count)]
|
| 467 |
+
width = max((len(row) for row in partials), default=0)
|
| 468 |
+
for index in range(width):
|
| 469 |
+
todo = [row for row in range(count) if index < len(partials[row])]
|
| 470 |
+
before = [running[row] for row in todo]
|
| 471 |
+
step_rights = [partials[row][index] for row in todo]
|
| 472 |
+
wrote = _write_lines(model, before, step_rights, scratch.OP_ADD, device)
|
| 473 |
+
for row, answer, left_tokens, right_tokens in zip(todo, wrote, before, step_rights):
|
| 474 |
+
pair = (scratch._read_number(lefts[row]), scratch._read_number(rights[row]))
|
| 475 |
+
if pair in watched:
|
| 476 |
+
print(
|
| 477 |
+
f" use {left_tokens}+{right_tokens}={answer} {scratch._read_number(left_tokens)}+{scratch._read_number(right_tokens)}={scratch._read_number(answer)}",
|
| 478 |
+
flush=True,
|
| 479 |
+
)
|
| 480 |
+
running[row] = answer
|
| 481 |
+
return running
|
| 482 |
+
|
| 483 |
+
|
| 484 |
+
def _report_power() -> None:
|
| 485 |
+
"""Score powers built by multiplying the written line by the base again."""
|
| 486 |
+
if not torch.cuda.is_available():
|
| 487 |
+
raise SystemExit("CUDA is not available. Refusing to build the model in system RAM.")
|
| 488 |
+
device = torch.device("cuda")
|
| 489 |
+
print(torch.cuda.get_device_name(0), flush=True)
|
| 490 |
+
model = _load(device)
|
| 491 |
+
bases = list(range(2, scratch.FAR_HI + 1))
|
| 492 |
+
limit = 10 ** scratch.WIDTH - 1
|
| 493 |
+
base_tokens = [scratch._number(base) for base in bases]
|
| 494 |
+
running = [list(tokens) for tokens in base_tokens]
|
| 495 |
+
previous = [list(tokens) for tokens in base_tokens]
|
| 496 |
+
square_hits = square_rows = 0
|
| 497 |
+
higher_hits = higher_rows = 0
|
| 498 |
+
shown = {2, 12, 37}
|
| 499 |
+
probe = {(12, 12), (144, 12)}
|
| 500 |
+
exponent = 1
|
| 501 |
+
while True:
|
| 502 |
+
todo = []
|
| 503 |
+
for index, base in enumerate(bases):
|
| 504 |
+
current = scratch._read_number(running[index])
|
| 505 |
+
if current is None or current < 0 or current * base > limit:
|
| 506 |
+
continue
|
| 507 |
+
todo.append(index)
|
| 508 |
+
if not todo:
|
| 509 |
+
break
|
| 510 |
+
exponent += 1
|
| 511 |
+
print(f"power exp {exponent} rows {len(todo)}", flush=True)
|
| 512 |
+
wrote = _multiply_written(
|
| 513 |
+
model,
|
| 514 |
+
[running[index] for index in todo],
|
| 515 |
+
[base_tokens[index] for index in todo],
|
| 516 |
+
device,
|
| 517 |
+
probe,
|
| 518 |
+
)
|
| 519 |
+
for index, tokens in zip(todo, wrote):
|
| 520 |
+
previous[index] = running[index]
|
| 521 |
+
running[index] = tokens
|
| 522 |
+
got = scratch._read_number(tokens)
|
| 523 |
+
expect = bases[index] ** exponent
|
| 524 |
+
if exponent == 2:
|
| 525 |
+
square_rows += 1
|
| 526 |
+
square_hits += int(got == expect)
|
| 527 |
+
else:
|
| 528 |
+
higher_rows += 1
|
| 529 |
+
higher_hits += int(got == expect)
|
| 530 |
+
if got != expect:
|
| 531 |
+
print(f"miss ^{exponent} {bases[index]} truth {expect} pred {got}", flush=True)
|
| 532 |
+
if bases[index] in shown:
|
| 533 |
+
print(f"power {bases[index]}^{exponent} = {got} tokens {tokens}", flush=True)
|
| 534 |
+
print(f"power squares {square_hits}/{square_rows}", flush=True)
|
| 535 |
+
print(f"power higher {higher_hits}/{higher_rows}", flush=True)
|
| 536 |
+
print("power divide-back", flush=True)
|
| 537 |
+
quotients = _write_lines(model, running, base_tokens, scratch.OP_DIV, device)
|
| 538 |
+
back_hits = 0
|
| 539 |
+
for index, tokens in enumerate(quotients):
|
| 540 |
+
got = scratch._read_number(tokens)
|
| 541 |
+
expect = scratch._read_number(previous[index])
|
| 542 |
+
if got == expect:
|
| 543 |
+
back_hits += 1
|
| 544 |
+
else:
|
| 545 |
+
print(f"miss /base {bases[index]} truth {expect} pred {got}", flush=True)
|
| 546 |
+
if bases[index] in shown:
|
| 547 |
+
print(
|
| 548 |
+
f"power {scratch._read_number(running[index])}/{bases[index]} = {got}",
|
| 549 |
+
flush=True,
|
| 550 |
+
)
|
| 551 |
+
print(f"power divide-back {back_hits}/{len(bases)}", flush=True)
|
| 552 |
+
|
| 553 |
+
|
| 554 |
def _chain(model: FluidForCausalLM, rows: list[tuple[int, int, int, int]], extra, device: torch.device) -> tuple[int, int]:
|
| 555 |
prompts = _prompts(rows, device)
|
| 556 |
first = _generate_answer(model, prompts)
|
|
|
|
| 680 |
if __name__ == "__main__":
|
| 681 |
if len(sys.argv) > 1 and sys.argv[1] == "think":
|
| 682 |
_report_thought()
|
| 683 |
+
elif len(sys.argv) > 1 and sys.argv[1] == "power":
|
| 684 |
+
_report_power()
|
| 685 |
else:
|
| 686 |
main()
|