Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75937.49433153236,
-      "required_pages_per_second_upper_estimate": 0.006582069546837413
+      "required_token_positions_per_second": 76190.58439574811,
+      "required_pages_per_second_upper_estimate": 0.0066040067521527315
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9492.186791441545,
-      "required_pages_per_second_upper_estimate": 0.0037865752319457256
+      "required_token_positions_per_second": 9523.823049468514,
+      "required_pages_per_second_upper_estimate": 0.0037991954082768923
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18984.37358288309,
-      "required_pages_per_second_upper_estimate": 0.0043917007148363395
+      "required_token_positions_per_second": 19047.646098937028,
+      "required_pages_per_second_upper_estimate": 0.004406337697867168
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28476.560374324636,
-      "required_pages_per_second_upper_estimate": 0.0017340731218307103
+      "required_token_positions_per_second": 28571.469148405544,
+      "required_pages_per_second_upper_estimate": 0.0017398525682243834
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28476.560374324636,
-      "required_pages_per_second_upper_estimate": 0.0021922044457798537
+      "required_token_positions_per_second": 28571.469148405544,
+      "required_pages_per_second_upper_estimate": 0.002199510785932904
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28476.560374324636,
-      "required_pages_per_second_upper_estimate": 0.35358357493232473
+      "required_token_positions_per_second": 28571.469148405544,
+      "required_pages_per_second_upper_estimate": 0.3547620242671759
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05602426998439469,
-    "producer_p95_seconds": 0.16608884800007218,
-    "gpu_update_seconds": 0.9709037761782602,
-    "producer_batches_per_second": 17.84940705659432,
-    "gpu_batches_per_second": 1.029968184835237,
-    "producer_headroom": 17.33005671379031,
+    "producer_batch_seconds": 0.06179190702345494,
+    "producer_p95_seconds": 0.17868348000001788,
+    "gpu_update_seconds": 0.9676786257084341,
+    "producer_batches_per_second": 16.183349052820468,
+    "gpu_batches_per_second": 1.0334009385273997,
+    "producer_headroom": 15.6602809707932,
     "sustainable": true,
-    "buffer_seconds": 60.19603412305213,
+    "buffer_seconds": 60.963753419631345,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -32,3 +32,4 @@
 {"step": 118200, "loss": 0.5931202527135611, "lr": 0.0003, "grad_norm": 0.1292053724103761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17296945848, "tokens_per_second": 150656.84809803465, "data_wait_s": 1.3708000096812611e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 118300, "loss": 0.6326055247336626, "lr": 0.0003, "grad_norm": 0.14419355422351368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17311707810, "tokens_per_second": 151942.82823085733, "data_wait_s": 1.4157000350678572e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 118400, "loss": 0.6232305392622948, "lr": 0.0003, "grad_norm": 0.14040777355668516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17326463478, "tokens_per_second": 151948.54826641432, "data_wait_s": 1.3704000139114214e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 118500, "loss": 0.5947024971246719, "lr": 0.0003, "grad_norm": 0.13812125208627182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17341101633, "tokens_per_second": 150676.36652273164, "data_wait_s": 1.222200035044807e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6486b21d03e34d239161d50faf4d05da2c3f5df509e281f5d19c874dcc92226
-size 72177
+oid sha256:93b1b9ad9e9e741f867cbadfc8fc01f3e2eb5b461b90491cb8318b50bb47db0e
+size 74295