-
Notifications
You must be signed in to change notification settings - Fork 4
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
1 changed file
with
2 additions
and
58 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -1,59 +1,3 @@ | ||
. | ||
various experiments for scaling inference time compute with small reasoning models | ||
|
||
|
||
|
||
``` | ||
curl -X 'POST' 'https://rawsh--vllm-qwen-serve.modal.run/v1/completions' -H 'accept: application/json' -H 'Authorization: Bearer 9FF74944EED19865193F979942FB1' -H 'Content-Type: application/json' -d '{ | ||
"model": "Qwen/Qwen2.5-0.5B-Instruct", | ||
"prompt": [ | ||
"<|im_start|>user\nIf four coins are tossed simultaneously, what is the probability of obtaining more heads than tails, in the coins that land face up? Express your answer as a common fraction.\n<|im_end|>\n<|im_start|>assistant\n## Step 1:\nIdentify all the possible outcomes of tossing four coins simultaneously. When tossing four coins simultaneously, each coin has 2 possible outcomes heads or tails. Therefore, for four coins, the total number of possible outcomes is $2^4 = 16$.\n\n## Step 2:\n" | ||
], | ||
"max_tokens": 200, | ||
"stop": ["\n\n## Step "], | ||
"temperature": 0.7 | ||
}' | ||
# one step | ||
curl -X 'POST' 'https://rawsh--vllm-gemma-serve.modal.run/v1/completions' -H 'accept: application/json' -H 'Authorization: Bearer 9FF74944EED19865193F979942FB1' -H 'Content-Type: application/json' -d '{ | ||
"model": "rawsh/mirrorgemma-2-2b-SFT", | ||
"prompt": [ | ||
"Find the least positive integer such that when its leftmost digit is deleted, the resulting integer is 1/29 of the original integer.\n\n" | ||
], | ||
"max_tokens": 200, | ||
"stop": ["\n\n"], | ||
"temperature": 0.3 | ||
}' | ||
``` | ||
|
||
https://huggingface.co/datasets/KbsdJames/Omni-MATH?row=0 | ||
|
||
A zoo plans to construct a new exhibit for endangered species. The exhibit area is divided into four sections, each housing a different species. The total area of the exhibit is 50,000 square feet, with the following area distribution: 40% for Species A, 30% for Species B, 20% for Species C, and 10% for Species D. Calculate the individual areas for each species' section and estimate the number of animals that can be housed in each section based on the recommended space requirements: 500 sq ft per animal for Species A, 400 sq ft per animal for Species B, 600 sq ft per animal for Species C, and 800 sq ft per animal for Species D. Nonenn1. Understand the problem: We need to calculate the individual areas for each species' section and estimate the number of animals that can be housed in each section based on the recommended space requirements. 2. Identify the total area of the exhibit: 50,000 sq ft. 3. Calculate the area distribution for each species: a. Species A: 40% of the total area b. Species B: 30% of the total area c. Species C: 20% of the total area d. Species D: 10% of the total area 4. Calculate the individual areas for each species' section: a. Species A: 50,000 sq ft * 40% = 20,000 sq ft b. Species B: 50,000 sq ft * 30% = 15,000 sq ft c. Species C: 50,000 sq ft * 20% = 10,000 sq ft d. Species D: 50,000 sq ft * 10% = 5,000 sq ft 5. Identify the recommended space requirements for each species: a. Species A: 500 sq ft per animal b. Species B: 400 sq ft per animal c. Species C: 600 sq ft per animal d. Species D: 800 sq ft per animal 6. Estimate the number of animals that can be housed in each section based on the recommended space requirements: a. Species A: 20,000 sq ft / 500 sq ft per animal = 40 animals b. Species B: 15,000 sq ft / 400 sq ft per animal = 37.5 animals round down to 37 animals c. Species C: 10,000 sq ft / 600 sq ft per animal = 16.67 animals round down to 16 animals d. Species D: 5,000 sq ft / 800 sq ft per animal = 6.25 animals round down to 6 animals 7. Summarize the results: The zoo can house approximately 40 animals of Species A, 37 animals of Species B, 16 animals of Species C, and 6 animals of Species D in their respective sections. | ||
|
||
|
||
## Step 1: The problem presents a rhombus with diagonals measuring 18 feet and 12 feet and asks for the perimeter of the rhombus. ## Step 2: To find the perimeter, we first recall that a rhombus is a type of quadrilateral where all four sides are of equal length. This property will be useful in our solution. ## Step 3: The diagonals of a rhombus bisect each other at right angles, creating four right-angled triangles. ## Step 4: Since the diagonals of the rhombus are perpendicular bisectors of each other, we can use this information to find the side length of the rhombus. ## Step 5: Considering the right-angled triangles formed by the diagonals, we can apply the Pythagorean theorem, which states that in a right-angled triangle, the square of the length of the hypotenuse is equal to the sum of the squares of the lengths of the other two sides. ## Step 6: The two legs of the right-angled triangle are $frac{18}{2}=9$ feet and $frac{12}{2}=6$ feet, so applying the Pythagorean theorem yields the hypotenuse length, which is a side of the rhombus. ## Step 7: Using the Pythagorean theorem formula $a^2 + b^2 = c^2$, where $c$ is the length of the hypotenuse, we can calculate $c = sqrt{9^2+6^2} = sqrt{81+36} = sqrt{117} = 3sqrt{13}$ feet. ## Step 8: With each side of the rhombus equal in length, the perimeter can be found by multiplying the length of one side by $4$, giving a perimeter of $4 times 3sqrt{13} = 12sqrt{13}$ feet. The final answer is: $boxed{12sqrt{13}}$ | ||
|
||
|
||
How can I modify the given HTML code to ensure that it adjusts its layout based on different screen sizes? None 1. The task is to modify the HTML code to make it responsive to different screen sizes. 2. To achieve this, CSS media queries can be used. Media queries allow us to apply different styles for different screen sizes. 3. First, I need to ensure that the outer container of the HTML code takes up the full width of the screen. This can be done by setting the width to 100% in the CSS. 4. Next, I need to create a new CSS class for the div that needs to adjust its width based on screen size. I'll call this class 'inner-container'. 5. The 'inner-container' class should initially have a maximum width of 400 pixels. This can be set using the 'max-width' property in the CSS. To center the 'inner-container' horizontally, I can use the 'margin: 0 auto' rule. 6. Now, I need to create a media query for screens with a maximum width of 600 pixels. This can be done using the '@media max-width: 600px' syntax in the CSS. The value of 600 pixels can be adjusted as needed. 7. Inside the media query, I need to override the 'max-width' property of the 'inner-container' class to '100%'. This ensures that the 'inner-container' takes up the full width of the screen when the screen size is 600 pixels or less. 8. Optionally, I can add some padding to the sides of the 'inner-container' to create some spacing. This can be done using the 'padding' property in the CSS. 9. By following these steps, the HTML code will be responsive and adapt its layout based on the screen size. This will provide a better user experience across different devices. | ||
|
||
optimize cot | ||
|
||
preference tuning on high quality and low quality chains? | ||
- use tree of thought and track which branches got correct answer | ||
- rate every correct branch using critic model | ||
|
||
gen thoughts | ||
- we want high diversity | ||
- possibly off topic or wrong thoughts is fine if we catch them | ||
- signal thought quality based on answer correctness | ||
- no tree of thoughts, just backtracking at inference time? | ||
|
||
magpie filter reasoning/math questions | ||
dspy | ||
- question -> chain of thoughts -> answer | ||
- reward based on | ||
- avg quality of thoughts | ||
- quality of answer | ||
- quality of entire output | ||
|
||
|
||
- optimize just for output quality | ||
high throughput async mcts implementation for policy + prm hosted on serverless gpus on modal |