Add cpu_colab

data/cpu_colab/bandwidth.csv

@@ -0,0 +1,4 @@
+op,size_mb,median_ms,p10_ms,p90_ms,gbs
+copy,64,0.689424499995539,0.555079999969621,0.7566400000200701,194.7
+add,64,1.5741994999700637,1.4337799999566414,1.7313800000238189,127.9
+sum,64,0.11289000002534522,0.11025999998537372,0.1293099999770675,594.5

data/cpu_colab/hardware.json

@@ -0,0 +1,15 @@
+{
+  "env": {
+    "timestamp": "2026-10-19T06:44:55+00:00",
+    "platform": "colab",
+    "torch": "2.9.0+cpu",
+    "gpu": "cpu",
+    "gpu_mem_gb": 0.0
+  },
+  "launch_overhead_us": 2.43,
+  "copy_gbs": 194.7,
+  "add_gbs": 127.9,
+  "fp32_tflops": 0.84,
+  "fp32_size_at_70pct": 1024,
+  "fp32_ridge_flop_per_byte": 4.3
+}

data/cpu_colab/matmul.csv

@@ -0,0 +1,4 @@
+dtype,m,k,n,kind,median_ms,p10_ms,p90_ms,tflops,intensity
+fp32,256,256,256,square,0.07615500001634246,0.07261000001790308,0.0891800000317744,0.441,42.7
+fp32,1024,1024,1024,square,2.7535099999909107,2.6930200000379045,2.81297000003633,0.78,170.7
+fp32,2048,2048,2048,square,20.460063000001583,12.368949000006069,20.68896800000175,0.84,341.3

data/cpu_colab/train_step.csv

@@ -0,0 +1,2 @@
+preset,seq_len,n_kv_head,window,batch,precision,ckpt,tag,attn_type,params,tokens,gpu,platform,timestamp,status,step_ms,p10_ms,p90_ms,tokens_per_s,peak_mem_gb,approx_tflops,sm_mhz,power_w,temp_c
+10m,128,4,0,2,fp32,False,,mha,16060928,256,cpu,colab,2026-10-19T06:44:55+00:00,ok,94.22,92.8,94.22,2717,0,0.26,0.0,0.0,0.0