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": 75341.36127254891,
-      "required_pages_per_second_upper_estimate": 0.006530398244170153
+      "required_token_positions_per_second": 75335.00870894277,
+      "required_pages_per_second_upper_estimate": 0.0065298476200571975
     },
     {
       "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": 9417.670159068613,
-      "required_pages_per_second_upper_estimate": 0.003756849433169225
+      "required_token_positions_per_second": 9416.876088617846,
+      "required_pages_per_second_upper_estimate": 0.0037565326665940026
     },
     {
       "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": 18835.340318137227,
-      "required_pages_per_second_upper_estimate": 0.0043572244919332604
+      "required_token_positions_per_second": 18833.752177235692,
+      "required_pages_per_second_upper_estimate": 0.004356857103486015
     },
     {
       "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": 28253.010477205844,
-      "required_pages_per_second_upper_estimate": 0.0017204601059718416
+      "required_token_positions_per_second": 28250.628265853535,
+      "required_pages_per_second_upper_estimate": 0.001720315041798983
     },
     {
       "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": 28253.010477205844,
-      "required_pages_per_second_upper_estimate": 0.0021749949558739273
+      "required_token_positions_per_second": 28250.628265853535,
+      "required_pages_per_second_upper_estimate": 0.0021748115666496435
     },
     {
       "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": 28253.010477205844,
-      "required_pages_per_second_upper_estimate": 0.3508078333834864
+      "required_token_positions_per_second": 28250.628265853535,
+      "required_pages_per_second_upper_estimate": 0.35077825429123927
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04785197319529999,
-    "producer_p95_seconds": 0.1573215110001911,
-    "gpu_update_seconds": 0.9785859819188487,
-    "producer_batches_per_second": 20.89777982443198,
-    "gpu_batches_per_second": 1.0218826127461604,
-    "producer_headroom": 20.450274389415675,
+    "producer_batch_seconds": 0.09414915059373996,
+    "producer_p95_seconds": 0.16845601699969848,
+    "gpu_update_seconds": 0.9786685003893548,
+    "producer_batches_per_second": 10.621444736289428,
+    "gpu_batches_per_second": 1.0217964505878738,
+    "producer_headroom": 10.394873392032782,
     "sustainable": true,
-    "buffer_seconds": 60.67233087896862,
+    "buffer_seconds": 60.67744702414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "upload_pages_per_commit": 162,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 611305.9510963782,
-    "upload_bytes_per_second": 5634180.224215531,
+    "generation_bytes_per_second": 606365.3590635646,
+    "upload_bytes_per_second": 5666132.37370981,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.58601331552748,
+    "estimated_full_shards_per_hour": 65.05594529595471,
     "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": [
@@ -340,6 +340,12 @@
       "bytes": 1090489,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.169713431999753,
+      "bytes": 7523243,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -29,3 +29,4 @@
 {"step": 84800, "loss": 0.6267599500715733, "lr": 0.0003, "grad_norm": 0.1399491145580503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12410053275, "tokens_per_second": 149362.03363209727, "data_wait_s": 1.4009000096848467e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84900, "loss": 0.6152340620756149, "lr": 0.0003, "grad_norm": 0.12822516601325828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12424713818, "tokens_per_second": 147683.3784253226, "data_wait_s": 6.062000011297641e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 85000, "loss": 0.6123912185430527, "lr": 0.0003, "grad_norm": 0.12991575957755994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12439396796, "tokens_per_second": 148490.2662518952, "data_wait_s": 1.2061000234098174e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 85100, "loss": 0.6527527756989002, "lr": 0.0003, "grad_norm": 0.1299015585553546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12454100277, "tokens_per_second": 134800.67630284632, "data_wait_s": 9.854999916569795e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8999f711a72bd394a6346638cd7879962e83bf1ba4f250068ba22720a8673666
-size 66606
+oid sha256:82a177f24055d8438d5870cf82b0289306de6f5aefd2c84d2d8a35f2e10dfa44
+size 68724