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": 76662.52684115656,
-      "required_pages_per_second_upper_estimate": 0.00530369148844078
+      "required_token_positions_per_second": 75358.14875426136,
+      "required_pages_per_second_upper_estimate": 0.005213451585815228
     },
     {
       "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": 9582.81585514457,
-      "required_pages_per_second_upper_estimate": 0.004012124886598488
+      "required_token_positions_per_second": 9419.76859428267,
+      "required_pages_per_second_upper_estimate": 0.003943860403289591
     },
     {
       "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": 19165.63171028914,
-      "required_pages_per_second_upper_estimate": 0.003991695653556218
+      "required_token_positions_per_second": 18839.53718856534,
+      "required_pages_per_second_upper_estimate": 0.003923778764371993
     },
     {
       "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": 28748.44756543371,
-      "required_pages_per_second_upper_estimate": 0.001723425535209999
+      "required_token_positions_per_second": 28259.30578284801,
+      "required_pages_per_second_upper_estimate": 0.0016941022322202427
     },
     {
       "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": 28748.44756543371,
-      "required_pages_per_second_upper_estimate": 0.0022350653471553904
+      "required_token_positions_per_second": 28259.30578284801,
+      "required_pages_per_second_upper_estimate": 0.0021970367250666754
     },
     {
       "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": 28748.44756543371,
-      "required_pages_per_second_upper_estimate": 0.36875421769132916
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28259.30578284801,
+      "required_pages_per_second_upper_estimate": 0.3624800321038469
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07185018067181659,
-    "producer_p95_seconds": 0.2559240359987598,
-    "gpu_update_seconds": 0.9617214959893398,
-    "producer_batches_per_second": 13.917849484159369,
-    "gpu_batches_per_second": 1.0398020676155133,
-    "producer_headroom": 13.38509502686021,
+    "producer_batch_seconds": 0.368102021453268,
+    "producer_p95_seconds": 0.27170388299782644,
+    "gpu_update_seconds": 0.978367983009015,
+    "producer_batches_per_second": 2.716638164745732,
+    "gpu_batches_per_second": 1.0221103075393523,
+    "producer_headroom": 2.6578718018075937,
     "sustainable": true,
-    "buffer_seconds": 60.58845424732841,
+    "buffer_seconds": 60.65881494655893,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 550794.1411728457,
-    "upload_bytes_per_second": 5504520.585807981,
+    "generation_bytes_per_second": 498670.82933374343,
+    "upload_bytes_per_second": 5506034.347321068,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.18760086430578,
+    "estimated_full_shards_per_hour": 107.003151512234,
     "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": 1.1291984180006693,
-      "bytes": 2752026,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.8884318259988504,
       "bytes": 6185512,
@@ -562,6 +556,12 @@
       "bytes": 5934822,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.7901811400006409,
+      "bytes": 1002646,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -207,3 +207,4 @@
 {"step": 505700, "loss": 0.6090046167373657, "lr": 0.0003, "grad_norm": 0.15855559229407937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74017608939, "tokens_per_second": 151303.1249323467, "data_wait_s": 9.109000529861078e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 505800, "loss": 0.6245326995849609, "lr": 0.0003, "grad_norm": 0.16180282557480538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74032118960, "tokens_per_second": 151104.85959689948, "data_wait_s": 1.8319999071536586e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 505900, "loss": 0.5876098871231079, "lr": 0.0003, "grad_norm": 0.1824659212107423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74046714407, "tokens_per_second": 151443.90136277847, "data_wait_s": 1.7674999980954453e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 506000, "loss": 0.5967216491699219, "lr": 0.0003, "grad_norm": 0.15872306353266838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74061387830, "tokens_per_second": 153078.95207051752, "data_wait_s": 1.205899752676487e-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 sha256:1a957608a39bcce9af8dbcbe667086e12baf411cf0967d43d05656ede17f9e76
-size 445959
+oid sha256:1ef151e29d71716a16f5cf8c43051e2268076b94d278808586f25699161fa195
+size 448077