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": 76413.05350459921,
-      "required_pages_per_second_upper_estimate": 0.0052864323441598506
+      "required_token_positions_per_second": 76150.50956321313,
+      "required_pages_per_second_upper_estimate": 0.005268268945108888
     },
     {
       "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": 9551.631688074902,
-      "required_pages_per_second_upper_estimate": 0.003999068727045876
+      "required_token_positions_per_second": 9518.81369540164,
+      "required_pages_per_second_upper_estimate": 0.003985328518831199
     },
     {
       "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": 19103.263376149804,
-      "required_pages_per_second_upper_estimate": 0.003978705974318571
+      "required_token_positions_per_second": 19037.62739080328,
+      "required_pages_per_second_upper_estimate": 0.003965035729508225
     },
     {
       "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": 28654.895064224704,
-      "required_pages_per_second_upper_estimate": 0.0017178172056089172
+      "required_token_positions_per_second": 28556.441086204923,
+      "required_pages_per_second_upper_estimate": 0.0017119150399571514
     },
     {
       "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": 28654.895064224704,
-      "required_pages_per_second_upper_estimate": 0.002227792051680355
+      "required_token_positions_per_second": 28556.441086204923,
+      "required_pages_per_second_upper_estimate": 0.0022201376879425997
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28654.895064224704,
-      "required_pages_per_second_upper_estimate": 0.3675542266546697
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28556.441086204923,
+      "required_pages_per_second_upper_estimate": 0.3662913647362774
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11514162374223247,
-    "producer_p95_seconds": 0.30869785899994895,
-    "gpu_update_seconds": 0.9648613243228449,
-    "producer_batches_per_second": 8.68495655609912,
-    "gpu_batches_per_second": 1.0364183689317386,
-    "producer_headroom": 8.37977868440417,
+    "producer_batch_seconds": 0.10760444438295735,
+    "producer_p95_seconds": 0.3218163010024,
+    "gpu_update_seconds": 0.9681878745512243,
+    "producer_batches_per_second": 9.293296440814878,
+    "gpu_batches_per_second": 1.0328573888239627,
+    "producer_headroom": 8.997656928607014,
     "sustainable": true,
-    "buffer_seconds": 60.786263432339226,
+    "buffer_seconds": 60.02764822217591,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313682.5314962577,
-    "upload_bytes_per_second": 5298031.726554063,
+    "generation_bytes_per_second": 323056.80236905173,
+    "upload_bytes_per_second": 5254465.453152432,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.308969103487,
+    "estimated_full_shards_per_hour": 69.32046941093124,
     "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": 1.2493900669978757,
-      "bytes": 6142029,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9163747099992179,
       "bytes": 2165474,
@@ -562,6 +556,12 @@
       "bytes": 16798507,
       "pages": 50,
       "shards": 4
+    },
+    {
+      "seconds": 1.2172141400005785,
+      "bytes": 2428421,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 637300, "loss": 0.556537926197052, "lr": 0.0001, "grad_norm": 0.1721248058661298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93284114297, "tokens_per_second": 147904.51063344016, "data_wait_s": 8.86399720911868e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 637400, "loss": 0.5920100212097168, "lr": 0.0001, "grad_norm": 0.1860952955176649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93298632099, "tokens_per_second": 150522.39252012363, "data_wait_s": 9.453000529902056e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 637500, "loss": 0.569724440574646, "lr": 0.0001, "grad_norm": 0.17819412745127783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93313262112, "tokens_per_second": 152721.54016342905, "data_wait_s": 1.1635002010734752e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 637600, "loss": 0.5447022914886475, "lr": 0.0001, "grad_norm": 0.21281934312340497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93328093606, "tokens_per_second": 152960.06271889122, "data_wait_s": 1.8478000129107386e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:783c94546a3fce40ffef70a8a64b52b1fe6ade3a17af8eb55654e786b843a110
-size 706704
+oid sha256:a0c46f68f04f2b12ad7a41c31bb84ccfda73609c46b8903b3c36edc76061490b
+size 708822