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": 76836.6045221097,
-      "required_pages_per_second_upper_estimate": 0.005315734586324987
+      "required_token_positions_per_second": 76639.29784399322,
+      "required_pages_per_second_upper_estimate": 0.005302084452518321
     },
     {
       "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": 9604.575565263713,
-      "required_pages_per_second_upper_estimate": 0.004021235222830955
+      "required_token_positions_per_second": 9579.912230499152,
+      "required_pages_per_second_upper_estimate": 0.004010909199594029
     },
     {
       "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": 19209.151130527425,
-      "required_pages_per_second_upper_estimate": 0.004000759601107563
+      "required_token_positions_per_second": 19159.824460998305,
+      "required_pages_per_second_upper_estimate": 0.0039904861566764
     },
     {
       "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": 28813.72669579114,
-      "required_pages_per_second_upper_estimate": 0.0017273389193994662
+      "required_token_positions_per_second": 28739.73669149746,
+      "required_pages_per_second_upper_estimate": 0.001722903331618255
     },
     {
       "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": 28813.72669579114,
-      "required_pages_per_second_upper_estimate": 0.0022401405124081314
+      "required_token_positions_per_second": 28739.73669149746,
+      "required_pages_per_second_upper_estimate": 0.002234388115021932
     },
     {
       "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": 28813.72669579114,
-      "required_pages_per_second_upper_estimate": 0.369591548284285
+      "required_token_positions_per_second": 28739.73669149746,
+      "required_pages_per_second_upper_estimate": 0.36864248395348265
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11805769182021209,
-    "producer_p95_seconds": 0.26477346400133683,
-    "gpu_update_seconds": 0.9595426614509598,
-    "producer_batches_per_second": 8.470434959230627,
-    "gpu_batches_per_second": 1.0421631472725383,
-    "producer_headroom": 8.127743704427408,
+    "producer_batch_seconds": 0.3162383477343269,
+    "producer_p95_seconds": 0.27239343600012944,
+    "gpu_update_seconds": 0.962012989081405,
+    "producer_batches_per_second": 3.1621718465342603,
+    "gpu_batches_per_second": 1.0394870041774256,
+    "producer_headroom": 3.0420503900734897,
     "sustainable": true,
-    "buffer_seconds": 60.451187671410466,
+    "buffer_seconds": 60.60681831212851,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418553.8117773817,
-    "upload_bytes_per_second": 5749121.842075235,
+    "generation_bytes_per_second": 380101.6551646966,
+    "upload_bytes_per_second": 5746295.051220811,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.81190457335556,
+    "estimated_full_shards_per_hour": 81.56096688466715,
     "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": 1.203395697997621,
-      "bytes": 7988495,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9280091689979599,
       "bytes": 6513685,
@@ -562,6 +556,12 @@
       "bytes": 6195866,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1199947279965272,
+      "bytes": 7293697,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 508400, "loss": 0.5920135974884033, "lr": 0.0003, "grad_norm": 0.149531166071458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74412898822, "tokens_per_second": 152571.57669540404, "data_wait_s": 8.951999916462228e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 508500, "loss": 0.5951050519943237, "lr": 0.0003, "grad_norm": 0.15760390383220357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74427372681, "tokens_per_second": 150714.5333926887, "data_wait_s": 1.4768000255571678e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 508600, "loss": 0.608340859413147, "lr": 0.0003, "grad_norm": 0.16196406301846425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74441825612, "tokens_per_second": 150456.84190610118, "data_wait_s": 1.2659002095460892e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 508700, "loss": 0.5609819889068604, "lr": 0.0003, "grad_norm": 0.14620061298802478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74456476855, "tokens_per_second": 153012.41598693808, "data_wait_s": 1.0538999049458653e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:392d146eecdb656c51bcab4c14502f14901396fa99c36d30678d21565d6890b2
-size 503145
+oid sha256:31cad6f74a67055ac4b3ca5beecc67e39e0396ee41f3b860732221bbb08ebb12
+size 505263