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": 76710.99599893138,
-      "required_pages_per_second_upper_estimate": 0.005307044697239582
+      "required_token_positions_per_second": 76674.99414922333,
+      "required_pages_per_second_upper_estimate": 0.005304554005740979
     },
     {
       "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": 9588.874499866422,
-      "required_pages_per_second_upper_estimate": 0.004014661514624638
+      "required_token_positions_per_second": 9584.374268652917,
+      "required_pages_per_second_upper_estimate": 0.0040127773617910575
     },
     {
       "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": 19177.748999732845,
-      "required_pages_per_second_upper_estimate": 0.00399421936539293
+      "required_token_positions_per_second": 19168.748537305833,
+      "required_pages_per_second_upper_estimate": 0.003992344806427539
     },
     {
       "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": 28766.62349959927,
-      "required_pages_per_second_upper_estimate": 0.0017245151547101798
+      "required_token_positions_per_second": 28753.12280595875,
+      "required_pages_per_second_upper_estimate": 0.001723705808740796
     },
     {
       "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": 28766.62349959927,
-      "required_pages_per_second_upper_estimate": 0.0022364784460892776
+      "required_token_positions_per_second": 28753.12280595875,
+      "required_pages_per_second_upper_estimate": 0.002235428826542002
     },
     {
       "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": 28766.62349959927,
-      "required_pages_per_second_upper_estimate": 0.3689873590590073
+      "required_token_positions_per_second": 28753.12280595875,
+      "required_pages_per_second_upper_estimate": 0.36881418665690213
     }
   ],
   "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": 0.06520484366396317,
-    "producer_p95_seconds": 0.22371553800076072,
-    "gpu_update_seconds": 0.9611138408504963,
-    "producer_batches_per_second": 15.336283990704068,
-    "gpu_batches_per_second": 1.040459472641756,
-    "producer_headroom": 14.739914810679565,
+    "producer_batch_seconds": 0.30356047371077466,
+    "producer_p95_seconds": 0.23479802000110794,
+    "gpu_update_seconds": 0.961565120650835,
+    "producer_batches_per_second": 3.2942365248539462,
+    "gpu_batches_per_second": 1.039971166303485,
+    "producer_headroom": 3.167622941473572,
     "sustainable": true,
-    "buffer_seconds": 60.55017197358127,
+    "buffer_seconds": 60.57860260100261,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 480343.5060871804,
-    "upload_bytes_per_second": 5132874.784734549,
+    "generation_bytes_per_second": 452518.9777771293,
+    "upload_bytes_per_second": 5053238.400545098,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.07053458177145,
+    "estimated_full_shards_per_hour": 97.10003852830323,
     "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": 1.278834224000093,
-      "bytes": 9834674,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1737758689996554,
       "bytes": 8190004,
@@ -562,6 +556,12 @@
       "bytes": 6216793,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8775070329993468,
+      "bytes": 1128371,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -155,3 +155,4 @@
 {"step": 455700, "loss": 0.5648539662361145, "lr": 0.0003, "grad_norm": 0.18789220262180192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66700443276, "tokens_per_second": 151517.46830350376, "data_wait_s": 1.917999907163903e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455800, "loss": 0.5343227386474609, "lr": 0.0003, "grad_norm": 0.15610936897271188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66715125888, "tokens_per_second": 151812.79188532886, "data_wait_s": 1.8380998881184496e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455900, "loss": 0.5676047801971436, "lr": 0.0003, "grad_norm": 0.16011803222362866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66729768871, "tokens_per_second": 151271.31900934165, "data_wait_s": 7.996000931598246e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 456000, "loss": 0.5559360384941101, "lr": 0.0003, "grad_norm": 0.19572494407885363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66744473858, "tokens_per_second": 153701.48077976104, "data_wait_s": 1.487400004407391e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:caece00592a31f2a652c2bd287f4db60ffb5d870721f552b7b47260725caf1d3
-size 335040
+oid sha256:6a2e67b454dc4652a17b2d23c9c48b33bea19ef3d20aa63c35201c285c47bb0b
+size 337158