Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76681.02312107726,
-      "required_pages_per_second_upper_estimate": 0.005304971104002977
+      "required_token_positions_per_second": 76626.37720339632,
+      "required_pages_per_second_upper_estimate": 0.005301190572621811
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9585.127890134658,
-      "required_pages_per_second_upper_estimate": 0.004013092887368057
+      "required_token_positions_per_second": 9578.29715042454,
+      "required_pages_per_second_upper_estimate": 0.004010232999293496
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19170.255780269315,
-      "required_pages_per_second_upper_estimate": 0.003992658725388162
+      "required_token_positions_per_second": 19156.59430084908,
+      "required_pages_per_second_upper_estimate": 0.003989813399502368
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28755.383670403975,
-      "required_pages_per_second_upper_estimate": 0.0017238413441121472
+      "required_token_positions_per_second": 28734.89145127362,
+      "required_pages_per_second_upper_estimate": 0.0017226128668650894
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28755.383670403975,
-      "required_pages_per_second_upper_estimate": 0.0022356045988081268
+      "required_token_positions_per_second": 28734.89145127362,
+      "required_pages_per_second_upper_estimate": 0.0022340114189064878
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28755.383670403975,
-      "required_pages_per_second_upper_estimate": 0.3688431865984784
+      "required_token_positions_per_second": 28734.89145127362,
+      "required_pages_per_second_upper_estimate": 0.3685803344143048
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08229618513283299,
-    "producer_p95_seconds": 0.24172019499928865,
-    "gpu_update_seconds": 0.961489518516015,
-    "producer_batches_per_second": 12.151231535045222,
-    "gpu_batches_per_second": 1.0400529394677365,
-    "producer_headroom": 11.683281758007249,
+    "producer_batch_seconds": 0.09204167287502685,
+    "producer_p95_seconds": 0.21008302500013087,
+    "gpu_update_seconds": 0.9621752024671231,
+    "producer_batches_per_second": 10.864643902743802,
+    "gpu_batches_per_second": 1.0393117567734962,
+    "producer_headroom": 10.453690946855712,
     "sustainable": true,
-    "buffer_seconds": 60.573839666508945,
+    "buffer_seconds": 60.617037755428754,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343585.4507602706,
-    "upload_bytes_per_second": 5601983.848156459,
+    "generation_bytes_per_second": 384831.43039040995,
+    "upload_bytes_per_second": 5682598.006113981,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.72543947321022,
+    "estimated_full_shards_per_hour": 82.57586654457306,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.1955943939992721,
-      "bytes": 921426,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.0380561770007262,
       "bytes": 6122044,
@@ -562,6 +556,12 @@
       "bytes": 11858220,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0470604990005086,
+      "bytes": 6246532,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -132,3 +132,4 @@
 {"step": 773200, "loss": 0.5310789346694946, "lr": 3e-05, "grad_norm": 0.25695695842821425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113174036181, "tokens_per_second": 151391.51501430312, "data_wait_s": 7.618000381626189e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 773300, "loss": 0.5370212197303772, "lr": 3e-05, "grad_norm": 0.2433236307142802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113188694398, "tokens_per_second": 152259.72881415344, "data_wait_s": 1.0412999472464435e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 773400, "loss": 0.5059680342674255, "lr": 3e-05, "grad_norm": 0.24319014548207935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113203388704, "tokens_per_second": 153087.1538654904, "data_wait_s": 1.8093000107910484e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 773500, "loss": 0.5205590128898621, "lr": 3e-05, "grad_norm": 0.25537978553874435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113218093707, "tokens_per_second": 152744.99197317872, "data_wait_s": 1.1875001291627996e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e41197d3eb13011715c9c105efe60494040925a8303250ac3e1c192b50f73336
-size 285543
+oid sha256:58e308aefb3fc644e62b4559e81cc045affed649702be5dab422217e58adb20f
+size 287661