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": 77398.49481506312,
-      "required_pages_per_second_upper_estimate": 0.005354607460556608
+      "required_token_positions_per_second": 77157.2423938553,
+      "required_pages_per_second_upper_estimate": 0.00533791705827471
     },
     {
       "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": 9674.81185188289,
-      "required_pages_per_second_upper_estimate": 0.004050641689337955
+      "required_token_positions_per_second": 9644.655299231912,
+      "required_pages_per_second_upper_estimate": 0.00403801577048342
     },
     {
       "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": 19349.62370376578,
-      "required_pages_per_second_upper_estimate": 0.004030016333602239
+      "required_token_positions_per_second": 19289.310598463824,
+      "required_pages_per_second_upper_estimate": 0.0040174547043313885
     },
     {
       "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": 29024.43555564867,
-      "required_pages_per_second_upper_estimate": 0.001739970593814125
+      "required_token_positions_per_second": 28933.965897695736,
+      "required_pages_per_second_upper_estimate": 0.001734547075959016
     },
     {
       "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": 29024.43555564867,
-      "required_pages_per_second_upper_estimate": 0.002256522199451728
+      "required_token_positions_per_second": 28933.965897695736,
+      "required_pages_per_second_upper_estimate": 0.0022494885814798578
     },
     {
       "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": 29024.43555564867,
-      "required_pages_per_second_upper_estimate": 0.37229429529699043
+      "required_token_positions_per_second": 28933.965897695736,
+      "required_pages_per_second_upper_estimate": 0.3711338476635207
     }
   ],
   "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": 0.08951467269525892,
-    "producer_p95_seconds": 0.31732402600027854,
-    "gpu_update_seconds": 0.9525766641349622,
-    "producer_batches_per_second": 11.171352917798965,
-    "gpu_batches_per_second": 1.0497842721227095,
-    "producer_headroom": 10.641570096311314,
+    "producer_batch_seconds": 0.11443244610131842,
+    "producer_p95_seconds": 0.3259402770054294,
+    "gpu_update_seconds": 0.9555551457327822,
+    "producer_batches_per_second": 8.738780250442261,
+    "gpu_batches_per_second": 1.0465120767395737,
+    "producer_headroom": 8.350386435738114,
     "sustainable": true,
-    "buffer_seconds": 60.01232984050262,
+    "buffer_seconds": 60.19997418116528,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318131.51268748654,
-    "upload_bytes_per_second": 4673379.252119518,
+    "generation_bytes_per_second": 323892.31378748885,
+    "upload_bytes_per_second": 4885618.457809836,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.26361690014312,
+    "estimated_full_shards_per_hour": 69.49975071161747,
     "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": 3.847735643001215,
-      "bytes": 2775913,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.592129067001224,
       "bytes": 15451622,
@@ -562,6 +556,12 @@
       "bytes": 870650,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.4489900529952138,
+      "bytes": 10666654,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -381,3 +381,4 @@
 {"step": 798100, "loss": 0.44999396800994873, "lr": 3e-05, "grad_norm": 0.26055396723796714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116818909689, "tokens_per_second": 151903.89110008214, "data_wait_s": 1.846699888119474e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 798200, "loss": 0.5165267586708069, "lr": 3e-05, "grad_norm": 0.269538701766825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116833409310, "tokens_per_second": 151597.68100082804, "data_wait_s": 9.444003808312118e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 798300, "loss": 0.5799460411071777, "lr": 3e-05, "grad_norm": 0.2847245988695983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116847986781, "tokens_per_second": 150896.43604142455, "data_wait_s": 1.353699917672202e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 798400, "loss": 0.5292727947235107, "lr": 3e-05, "grad_norm": 0.26238272934409546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116862567200, "tokens_per_second": 151617.72200572313, "data_wait_s": 8.610004442743957e-06, "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:ae8390c599a003a3b5e5bf53528e01c82dae27c04061f814b838d613ac50e025
-size 816840
+oid sha256:453069b8c7fd04b4dc4677ea5dea9bc31fb7ada9223305dae14c30023cf902a3
+size 818958