diff --git a/scripts/benchmarks/results/stats/cb_sync_smoke.json b/scripts/benchmarks/results/stats/cb_sync_smoke.json
deleted file mode 100644
index e8eda49452..0000000000
--- a/scripts/benchmarks/results/stats/cb_sync_smoke.json
+++ /dev/null
@@ -1,15 +0,0 @@
-{
- "backend": "cb_sync_driver",
- "use_cuda_graph": false,
- "attn_impl": "paged_attention",
- "n_prompts": 8,
- "n_decoded_tokens": 512,
- "wall_times_s": [
- 100.8571443540277,
- 100.87157070200192
- ],
- "median_wall_s": 100.87157070200192,
- "decode_tps": 5.075761152887835,
- "max_new_tokens": 64,
- "peak_memory_gb": 45.91296434402466
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/cb_tpaged_64_lora_4bit.json b/scripts/benchmarks/results/stats/cb_tpaged_64_lora_4bit.json
deleted file mode 100644
index f742fc3734..0000000000
--- a/scripts/benchmarks/results/stats/cb_tpaged_64_lora_4bit.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
- "backend": "tpaged",
- "lora_adapter": "outputs/lora_rank32_fresh",
- "attn_impl": "sdpa",
- "persistent_cb": false,
- "n_prompts": 64,
- "n_prompt_tokens": 9129,
- "n_decoded_tokens": 32768,
- "wall_times_s": [
- 60.9150581190479,
- 56.917108469991945,
- 58.08906611002749
- ],
- "median_wall_s": 58.08906611002749,
- "prompt_tps": 157.15522061774251,
- "decode_tps": 564.0992736556235,
- "max_new_tokens": 512,
- "sample_completions": [
- "FirstFirst??? ? ",
- "Let 11111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111",
- "First.$^ "
- ],
- "peak_memory_gb": 124.05089378356934,
- "sampling": {
- "temperature": 0.1,
- "top_p": 0.97,
- "min_p": 0.5,
- "top_k": 5
- }
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/cb_tpaged_64_lora_4bit_tied.json b/scripts/benchmarks/results/stats/cb_tpaged_64_lora_4bit_tied.json
deleted file mode 100644
index 3007331af5..0000000000
--- a/scripts/benchmarks/results/stats/cb_tpaged_64_lora_4bit_tied.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
- "backend": "tpaged",
- "lora_adapter": "outputs/lora_rank32_fresh",
- "attn_impl": "sdpa",
- "persistent_cb": false,
- "n_prompts": 64,
- "n_prompt_tokens": 9129,
- "n_decoded_tokens": 32275,
- "wall_times_s": [
- 61.436025188013446,
- 63.98777190799592,
- 65.29973084997619
- ],
- "median_wall_s": 63.98777190799592,
- "prompt_tps": 142.66788368762124,
- "decode_tps": 504.39324635973,
- "max_new_tokens": 512,
- "sample_completions": [
- "First list list list list list list list list list list list list list list list list list<|endoftext|>",
- "FirstFirst's???. \n,?.. and and and2222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222",
- "First. and. and and2222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222"
- ],
- "peak_memory_gb": 124.05089378356934,
- "sampling": {
- "temperature": 0.1,
- "top_p": 0.97,
- "min_p": 0.5,
- "top_k": 5
- }
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_256x512.json b/scripts/benchmarks/results/stats/flex_256x512.json
deleted file mode 100644
index ee0a9b09fc..0000000000
--- a/scripts/benchmarks/results/stats/flex_256x512.json
+++ /dev/null
@@ -1,23 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": true,
- "lora_adapter": null,
- "n_prompts": 256,
- "n_decoded_tokens": 117564,
- "wall_times_s": [
- 19.21345060999738,
- 16.626979330030736,
- 16.620332353981212
- ],
- "median_wall_s": 16.626979330030736,
- "best_wall_s": 16.620332353981212,
- "decode_tps_median": 7070.6769802535555,
- "decode_tps_best": 7073.504758876791,
- "max_new_tokens": 512,
- "peak_memory_gb": 154.22740983963013,
- "sample_completions": [
- "There are 7 choices for each of the 4 slots, so the total number of secret codes is $7^4 = 2401$.2401",
- "Let $h$ be the height of the tetrahedron. Then the volume of the tetrahedron is $\\frac{1}{3} \\cdot 120 \\cdot h = 40h$.400",
- "There are 3 choices for the color of the top triangle. For each choice of the top triangle, there are 2 choices for the color of the left triangle, and 2 choices for the color of the right triangle. Therefore, there are $3 \\times 2 \\times 2 = 12$ ways to color the triforce.1"
- ]
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_32x512_eager.json b/scripts/benchmarks/results/stats/flex_32x512_eager.json
deleted file mode 100644
index 21d807498c..0000000000
--- a/scripts/benchmarks/results/stats/flex_32x512_eager.json
+++ /dev/null
@@ -1,15 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": false,
- "lora_adapter": null,
- "n_prompts": 32,
- "n_decoded_tokens": 14922,
- "wall_times_s": [
- 40.0974782659905,
- 28.064613271970302
- ],
- "median_wall_s": 40.0974782659905,
- "decode_tps": 372.1431033895316,
- "max_new_tokens": 512,
- "peak_memory_gb": 43.89091157913208
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_32x512_mauto_nocg.json b/scripts/benchmarks/results/stats/flex_32x512_mauto_nocg.json
deleted file mode 100644
index c9d969c7da..0000000000
--- a/scripts/benchmarks/results/stats/flex_32x512_mauto_nocg.json
+++ /dev/null
@@ -1,23 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": true,
- "lora_adapter": null,
- "n_prompts": 32,
- "n_decoded_tokens": 14777,
- "wall_times_s": [
- 17.169199601979926,
- 6.7664765429799445,
- 6.7598927119979635
- ],
- "median_wall_s": 6.7664765429799445,
- "best_wall_s": 6.7598927119979635,
- "decode_tps_median": 2183.8544634180075,
- "decode_tps_best": 2185.9814392871463,
- "max_new_tokens": 512,
- "peak_memory_gb": 44.70592927932739,
- "sample_completions": [
- "First, we need to find the total number of letters in the word \"FLUFFY\". There are 6 letters in total. \n\nNext, we need to find the number of distinct arrangements of these 6 letters. Since there are 6 letters, the total number of arrangements is 6! (6 factorial), which is equal to 6 x 5 x 4 x 3",
- " \nTo determine the number of pairs of parallel edges in a cube, we need to consider the structure of the cube and the properties of its edges. A cube has 12 edges, and each edge is parallel to three other edges. However, we need to count each pair of parallel edges only once.\n\nLet's label the vertices of the cube as follows:\n- \\(A, B, C",
- "Let's denote the birth years of the two mathematicians as X and Y, where X and Y are uniformly distributed between 0 and 500. We want to find the probability that the two mathematicians were contemporaries for any length of time, which means that the difference between their birth years is less than or equal to 100 years.\n\nWe can visualize this problem as a"
- ]
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_64_lora_4bit.json b/scripts/benchmarks/results/stats/flex_64_lora_4bit.json
deleted file mode 100644
index 9e17e5537a..0000000000
--- a/scripts/benchmarks/results/stats/flex_64_lora_4bit.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": true,
- "lora_adapter": "outputs/lora_rank32_fresh",
- "n_prompts": 64,
- "n_decoded_tokens": 28172,
- "wall_times_s": [
- 19.392819664964918,
- 14.716534855018836,
- 14.195494230021723,
- 15.050612487946637,
- 16.052564749028534
- ],
- "median_wall_s": 15.050612487946637,
- "best_wall_s": 14.195494230021723,
- "decode_tps_median": 1871.8175105871403,
- "decode_tps_best": 1984.5733824765107,
- "max_new_tokens": 512,
- "peak_memory_gb": 40.59123468399048,
- "sample_completions": [
- "Let's denote the length of segment $DB$ as $ units.\n\nTo solve this problem, we can use the Power of a Point theorem, which states that for a point P inside a circle, the product of the lengths of the segments of any two intersecting chords through P is constant. In this case, we have two intersecting chords: AB and CD. Let",
- "Let $P(x)$ be a monic polynomial of degree $2023$ such that $P(k) = k^{2023}P(1-\\frac{1}{k})$ for every positive integer $1 \\leq k \\leq 2023$. We want to find $P(-1)$.\n\nFirst, we can rewrite the given equation",
- "First, let's consider the cube's edges. A cube has 12 edges. Each edge is parallel to 3 other edges. However, we need to be careful not to double-count the pairs.\n\nLet's count the pairs of parallel edges:\n\n1. Each edge is parallel to 3 other edges, so there are 12 * 3 = 36 pairs.\n2."
- ]
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_64_lora_4bit_tied.json b/scripts/benchmarks/results/stats/flex_64_lora_4bit_tied.json
deleted file mode 100644
index 9ade340295..0000000000
--- a/scripts/benchmarks/results/stats/flex_64_lora_4bit_tied.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": true,
- "lora_adapter": "outputs/lora_rank32_fresh",
- "n_prompts": 64,
- "n_decoded_tokens": 27691,
- "wall_times_s": [
- 19.849357629951555,
- 16.3931127290125,
- 15.936416806012858,
- 17.375963038997725,
- 17.99713112297468
- ],
- "median_wall_s": 17.375963038997725,
- "best_wall_s": 15.936416806012858,
- "decode_tps_median": 1593.638288585889,
- "decode_tps_best": 1737.59260548156,
- "max_new_tokens": 512,
- "peak_memory_gb": 40.59123468399048,
- "sample_completions": [
- "First, let's consider the cube's edges. A cube has 12 edges. Each edge is parallel to 3 other edges. However, we need to be careful not to double-count the pairs.\n\nLet's count the pairs of parallel edges:\n\n1. Each edge is parallel to 3 other edges, so there are 12 * 3 = 36 pairs.\n2.",
- "Let the common ratio of the geometric sequence be $r$. Then the second term is $\\frac{3}{4}r=15$, so $r=20$. The $n$th term of the sequence is $\\frac{3}{4}r^{n-1}$. We want to find the smallest $n$ such that $\\frac{3}{4}r^{",
- "Let $n = 20k + r$ and $n = 16m + s$, where $0 \\leq r < 20$ and $0 \\leq s < 16$. We want to find the number of integers $n$ such that $r < s$.\n\nSince $n$ is an integer, we have $20k +"
- ]
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_64_lora_bf16_fusedmerge.json b/scripts/benchmarks/results/stats/flex_64_lora_bf16_fusedmerge.json
deleted file mode 100644
index d855c9aedc..0000000000
--- a/scripts/benchmarks/results/stats/flex_64_lora_bf16_fusedmerge.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": true,
- "lora_adapter": "outputs/lora_rank32_fresh",
- "n_prompts": 64,
- "n_decoded_tokens": 29783,
- "wall_times_s": [
- 8.707227781007532,
- 7.633466306026094,
- 5.8300725820008665,
- 5.700851961970329,
- 5.8895474600140005
- ],
- "median_wall_s": 5.8895474600140005,
- "best_wall_s": 5.700851961970329,
- "decode_tps_median": 5056.92503578691,
- "decode_tps_best": 5224.306857760676,
- "max_new_tokens": 512,
- "peak_memory_gb": 52.00764560699463,
- "sample_completions": [
- " To solve this problem, we need to find the maximum value of \\(a\\) such that the line \\(y = mx + 2\\) does not pass through any lattice points for \\(0 < x \\leq 100\\) when \\(\\frac{1}{2} < m < a\\).\n\nFirst, let's consider the condition that the line \\(y = mx + 2",
- "First, we need to find the range of values for $x$ and $y$ in the set $S$. We are given that $\\frac{\\sqrt{2}}{2} \\le x \\le \\frac{\\sqrt{3}}{2}$. Since $x$ is a real number, we can write $x = \\cos \\theta$ for some angle $\\theta$.",
- "First, let's represent the given number in a more manageable form. The number \\(1\\underbrace{00\\ldots 0}_{100\\text{ zeros}}1\\underbrace{00\\ldots 0}_{100\\text{ zeros}}1\\) can be written as \\(10^{201} + 10^{10"
- ]
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_64_lora_bf16_mergeadapter.json b/scripts/benchmarks/results/stats/flex_64_lora_bf16_mergeadapter.json
deleted file mode 100644
index ca30b4afb3..0000000000
--- a/scripts/benchmarks/results/stats/flex_64_lora_bf16_mergeadapter.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": true,
- "lora_adapter": "outputs/lora_rank32_fresh",
- "n_prompts": 64,
- "n_decoded_tokens": 29357,
- "wall_times_s": [
- 7.899581302946899,
- 5.07448300399119,
- 6.057169139035977,
- 6.930038888996933,
- 5.760280788992532
- ],
- "median_wall_s": 6.057169139035977,
- "best_wall_s": 5.07448300399119,
- "decode_tps_median": 4846.653498712153,
- "decode_tps_best": 5785.219888786717,
- "max_new_tokens": 512,
- "peak_memory_gb": 44.46234083175659,
- "sample_completions": [
- " To solve this problem, we need to find the area of the region inside the larger circle \\( C \\) with radius 30 and outside the six smaller congruent circles that form a ring and are each internally tangent to \\( C \\).\n\nFirst, let's denote the radius of each of the six smaller circles as \\( r \\). Since the six smaller circles are congruent and form a ring",
- " \nWe know that the total number of pieces used to create an eight-row triangle is 15. We can set up an equation to represent the total number of pieces used.\n\nLet's denote the length of the shorter side of the triangle as x. Since the triangle has eight rows, the length of the longer side of the triangle will be 2x. \n\nThe total number of pieces used",
- "First, let's find the sum of the smallest and largest 2-digit prime numbers, which are 11 and 97, respectively. The sum is 11 + 97 = 108. Now, let's find the sum of the smallest and largest 2-digit prime numbers: $11 + 97 = 108$. Since 1"
- ]
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_64_lora_bf16_nomerge.json b/scripts/benchmarks/results/stats/flex_64_lora_bf16_nomerge.json
deleted file mode 100644
index 29001a9edd..0000000000
--- a/scripts/benchmarks/results/stats/flex_64_lora_bf16_nomerge.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
- "backend": "qwen3_flex",
- "capture_cudagraph": true,
- "lora_adapter": "outputs/lora_rank32_fresh",
- "n_prompts": 64,
- "n_decoded_tokens": 30064,
- "wall_times_s": [
- 15.238620115967933,
- 13.535559127980378,
- 13.255199212988373,
- 11.203266007010825,
- 14.028489015006926
- ],
- "median_wall_s": 13.535559127980378,
- "best_wall_s": 11.203266007010825,
- "decode_tps_median": 2221.112531498786,
- "decode_tps_best": 2683.503183909623,
- "max_new_tokens": 512,
- "peak_memory_gb": 45.04522657394409,
- "sample_completions": [
- " \nTo find the area of the smaller square, we need to determine its side length. Let's denote the side length of the smaller square as \\( s \\).\n\nFrom the diagram, we can see that the larger square has a side length of 6. The smaller square is inscribed within the larger square such that its vertices touch the midpoints of the sides of the larger square. This means",
- "First, let's find the time it takes for the first car to travel half the distance between $P_1$ and $P_2$. Since the distance between the two points is 600 miles, half the distance is 300 miles. The first car travels at a speed of 50 mph, so it will take 300 miles / 50",
- " \nA 10-digit palindrome has the form \\( \\overline{abcdeedcba} \\), where \\( a, b, c, d, e \\) are digits and \\( a \\neq 0 \\) (since it is a 10-digit number). The number can be expressed as:\n\n\\[\nN = 1000000000"
- ]
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/flex_verify_fusedmerge.json b/scripts/benchmarks/results/stats/flex_verify_fusedmerge.json
deleted file mode 100644
index 35b8a35306..0000000000
--- a/scripts/benchmarks/results/stats/flex_verify_fusedmerge.json
+++ /dev/null
@@ -1,11 +0,0 @@
-{
- "mode": "verify_no_drift",
- "n_iters": 10,
- "noise_scale": 0.01,
- "base_hash_before": "4ed669a411e7fc6d99ab431561fc789b1d070d1b5683ed9b8f9a8b7bfeb19924",
- "base_hash_after": "4ed669a411e7fc6d99ab431561fc789b1d070d1b5683ed9b8f9a8b7bfeb19924",
- "base_bit_identical": true,
- "inference_hash_initial_merged": "7a481386b1ee93dc4893e634aa604dd839dd3b252c931013226e3cf9506bf087",
- "inference_hash_after_restore": "7a481386b1ee93dc4893e634aa604dd839dd3b252c931013226e3cf9506bf087",
- "inference_deterministic": true
-}
\ No newline at end of file
diff --git a/scripts/benchmarks/results/stats/unsloth_fi_true_64_lora_4bit.json b/scripts/benchmarks/results/stats/unsloth_fi_true_64_lora_4bit.json
deleted file mode 100644
index 1fa2185459..0000000000
--- a/scripts/benchmarks/results/stats/unsloth_fi_true_64_lora_4bit.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
- "backend": "vllm",
- "lora_adapter": "outputs/lora_rank32_fresh",
- "n_prompts": 64,
- "n_prompt_tokens": 9129,
- "n_decoded_tokens": 30050,
- "wall_times_s": [
- 6.68179759796476,
- 6.6636974540306255,
- 6.6514031600090675,
- 6.653628617990762,
- 6.65495745599037
- ],
- "median_wall_s": 6.65495745599037,
- "prompt_tps": 1371.7593328538342,
- "decode_tps": 4515.430819614166,
- "max_new_tokens": 512,
- "sample_completions": [
- "First, we need to find the length of the legs of the trapezoid. Since the trapezoid is isosceles, the legs are equal in length. Let's call the length of each leg $x$. We can use the Pythagorean theore",
- "Let $P(x)$ be a monic polynomial of degree $2023$ such that $P(k) = k^{2023}P(1-\\frac{1}{k})$ for every positive integer $1 \\leq k \\leq 2023$. We want to find $P(-1)$.\n\nFirst, we can rewrite the given",
- "First, we need to find the value of $a$ such that the graph of $y = mx + 2$ passes through no lattice point with $0 < x \\leq 100$ for all $m$ such that $\\frac{1}{2} < m < a$.\n\nLet's consider the equat"
- ],
- "peak_memory_gb": 159.28503799438477,
- "sampling": {
- "temperature": 0.1,
- "top_p": 0.97,
- "min_p": 0.5,
- "top_k": 5
- }
-}
\ No newline at end of file