Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75723.0115804251,
-      "required_pages_per_second_upper_estimate": 0.006563478725572999
+      "required_token_positions_per_second": 75545.94438726809,
+      "required_pages_per_second_upper_estimate": 0.006548130990042847
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9465.376447553137,
-      "required_pages_per_second_upper_estimate": 0.003775880184918277
+      "required_token_positions_per_second": 9443.24304840851,
+      "required_pages_per_second_upper_estimate": 0.0037670508410756784
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18930.752895106274,
-      "required_pages_per_second_upper_estimate": 0.004379296512411046
+      "required_token_positions_per_second": 18886.48609681702,
+      "required_pages_per_second_upper_estimate": 0.00436905616769587
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28396.129342659413,
-      "required_pages_per_second_upper_estimate": 0.0017291752939913084
+      "required_token_positions_per_second": 28329.729145225534,
+      "required_pages_per_second_upper_estimate": 0.0017251318703424973
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28396.129342659413,
-      "required_pages_per_second_upper_estimate": 0.002186012642314919
+      "required_token_positions_per_second": 28329.729145225534,
+      "required_pages_per_second_upper_estimate": 0.002180900971308948
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28396.129342659413,
-      "required_pages_per_second_upper_estimate": 0.3525848907043895
+      "required_token_positions_per_second": 28329.729145225534,
+      "required_pages_per_second_upper_estimate": 0.351760422479426
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.15124620003138034,
-    "producer_p95_seconds": 0.37854810400313,
-    "gpu_update_seconds": 0.973653826772246,
-    "producer_batches_per_second": 6.611736359607854,
-    "gpu_batches_per_second": 1.0270590763404013,
-    "producer_headroom": 6.437542408141386,
+    "producer_batch_seconds": 0.12097615277326668,
+    "producer_p95_seconds": 0.39340934099891456,
+    "gpu_update_seconds": 0.975935910232999,
+    "producer_batches_per_second": 8.26609192866464,
+    "gpu_batches_per_second": 1.0246574488290485,
+    "producer_headroom": 8.06717595047097,
     "sustainable": true,
-    "buffer_seconds": 60.36653725987925,
+    "buffer_seconds": 60.50802643444594,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406386.5938844186,
-    "upload_bytes_per_second": 3781637.11005544,
+    "generation_bytes_per_second": 422489.17234266445,
+    "upload_bytes_per_second": 3775229.4387472877,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.20110285186213,
+    "estimated_full_shards_per_hour": 90.6563413401599,
     "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": 0.8130094469997857,
-      "bytes": 5715102,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8107168300011836,
       "bytes": 6329429,
@@ -562,6 +556,12 @@
       "bytes": 11452061,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.0961634519990184,
+      "bytes": 6066738,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 37800, "loss": 0.7046154029667377, "lr": 0.0003, "grad_norm": 0.155621582148506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5531510173, "tokens_per_second": 149407.14831093606, "data_wait_s": 1.4583994925487787e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37900, "loss": 0.6656279861927032, "lr": 0.0003, "grad_norm": 0.14468059415255227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5546372380, "tokens_per_second": 151590.1060227946, "data_wait_s": 1.3238000974524766e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 38000, "loss": 0.6564249563962221, "lr": 0.0003, "grad_norm": 0.13602110274995644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5560882393, "tokens_per_second": 147321.59861726186, "data_wait_s": 1.0757001291494817e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 38100, "loss": 0.5695996237918735, "lr": 0.0003, "grad_norm": 0.1426385781318986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5575418628, "tokens_per_second": 147514.8865444126, "data_wait_s": 1.2013995728921145e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c159b1305b9b0574263c6c1ba74b4ede42683ff31640f31f1b38e7e929d2234a
-size 804712
+oid sha256:5e11e8d5defc9e7dc36d02ec0bfd686959eb092068e65a57fe8a262d19a1bf3a
+size 806830