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": 75952.0772054699,
-      "required_pages_per_second_upper_estimate": 0.005254540934173707
+      "required_token_positions_per_second": 75414.2534814217,
+      "required_pages_per_second_upper_estimate": 0.0052173330410210834
     },
     {
       "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": 9494.009650683738,
-      "required_pages_per_second_upper_estimate": 0.003974943583275167
+      "required_token_positions_per_second": 9426.781685177713,
+      "required_pages_per_second_upper_estimate": 0.003946796637997354
     },
     {
       "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": 18988.019301367476,
-      "required_pages_per_second_upper_estimate": 0.003954703672732041
+      "required_token_positions_per_second": 18853.563370355427,
+      "required_pages_per_second_upper_estimate": 0.003926700048143579
     },
     {
       "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": 28482.028952051212,
-      "required_pages_per_second_upper_estimate": 0.0017074541461353865
+      "required_token_positions_per_second": 28280.34505553314,
+      "required_pages_per_second_upper_estimate": 0.0016953635044926068
     },
     {
       "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": 28482.028952051212,
-      "required_pages_per_second_upper_estimate": 0.002214352471816536
+      "required_token_positions_per_second": 28280.34505553314,
+      "required_pages_per_second_upper_estimate": 0.0021986724359759637
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28482.028952051212,
-      "required_pages_per_second_upper_estimate": 0.36533688577687834
+      "required_token_positions_per_second": 28280.34505553314,
+      "required_pages_per_second_upper_estimate": 0.3627499013036408
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1976480122970372,
-    "producer_p95_seconds": 0.27304690500022843,
-    "gpu_update_seconds": 0.9707173616930423,
-    "producer_batches_per_second": 5.059499401881869,
-    "gpu_batches_per_second": 1.0301659777217598,
-    "producer_headroom": 4.911343910882294,
+    "producer_batch_seconds": 0.12065899533587299,
+    "producer_p95_seconds": 0.2596936049994838,
+    "gpu_update_seconds": 0.9776401223432237,
+    "producer_batches_per_second": 8.28781971220915,
+    "gpu_batches_per_second": 1.0228712766034844,
+    "producer_headroom": 8.102505077402734,
     "sustainable": true,
-    "buffer_seconds": 60.18447642496862,
+    "buffer_seconds": 60.61368758527987,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375611.2547084938,
-    "upload_bytes_per_second": 4914260.644499286,
+    "generation_bytes_per_second": 376405.7676728683,
+    "upload_bytes_per_second": 4824617.76304396,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.59743147793876,
+    "estimated_full_shards_per_hour": 80.7679154647783,
     "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": 1.3504138789976423,
-      "bytes": 6609036,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9880936410008871,
       "bytes": 2162409,
@@ -562,6 +556,12 @@
       "bytes": 6177604,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 3.299507163999806,
+      "bytes": 8285983,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -265,3 +265,4 @@
 {"step": 786500, "loss": 0.5489715337753296, "lr": 3e-05, "grad_norm": 0.256187664183011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115121285169, "tokens_per_second": 152298.42278352476, "data_wait_s": 1.4785000530537218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786600, "loss": 0.5553094744682312, "lr": 3e-05, "grad_norm": 0.28255089079386364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115135798880, "tokens_per_second": 150367.19231246604, "data_wait_s": 9.563998901285231e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 786700, "loss": 0.5412750840187073, "lr": 3e-05, "grad_norm": 0.27167565713649355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115150412062, "tokens_per_second": 151988.70076290454, "data_wait_s": 1.2495002010837197e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 786800, "loss": 0.48189279437065125, "lr": 3e-05, "grad_norm": 0.24508727685326637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115165046750, "tokens_per_second": 151457.01707543465, "data_wait_s": 1.237800097442232e-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 sha256:1993df22c349eb822d5a5b389d3c8b6f80bd5bde57b2e0aeb90f62339635c288
-size 569586
+oid sha256:3587a733ceff763dcfaf489399c19dbc3ae4877fddf0df2ce46c1a0e65fe5eee
+size 571704