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": 77519.68628898173,
-      "required_pages_per_second_upper_estimate": 0.005362991767925259
+      "required_token_positions_per_second": 77061.43612280612,
+      "required_pages_per_second_upper_estimate": 0.005331288958142352
     },
     {
       "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": 9689.960786122716,
-      "required_pages_per_second_upper_estimate": 0.0040569842317584506
+      "required_token_positions_per_second": 9632.679515350765,
+      "required_pages_per_second_upper_estimate": 0.004033001759855189
     },
     {
       "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": 19379.921572245432,
-      "required_pages_per_second_upper_estimate": 0.004036326580598027
+      "required_token_positions_per_second": 19265.35903070153,
+      "required_pages_per_second_upper_estimate": 0.004012466224411821
     },
     {
       "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": 29069.882358368148,
-      "required_pages_per_second_upper_estimate": 0.001742695060243908
+      "required_token_positions_per_second": 28898.038546052296,
+      "required_pages_per_second_upper_estimate": 0.0017323932860858812
     },
     {
       "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": 29069.882358368148,
-      "required_pages_per_second_upper_estimate": 0.002260055488463808
+      "required_token_positions_per_second": 28898.038546052296,
+      "required_pages_per_second_upper_estimate": 0.0022466953879172994
     },
     {
       "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": 29069.882358368148,
-      "required_pages_per_second_upper_estimate": 0.37287723808530093
+      "required_token_positions_per_second": 28898.038546052296,
+      "required_pages_per_second_upper_estimate": 0.3706730101724233
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10616606402352602,
-    "producer_p95_seconds": 0.19707775800088712,
-    "gpu_update_seconds": 0.9510874402297387,
-    "producer_batches_per_second": 9.419205743356969,
-    "gpu_batches_per_second": 1.0514280366886628,
-    "producer_headroom": 8.958488279446632,
+    "producer_batch_seconds": 0.3238989407187063,
+    "producer_p95_seconds": 0.19328760700045677,
+    "gpu_update_seconds": 0.9567431352110554,
+    "producer_batches_per_second": 3.08738274284281,
+    "gpu_batches_per_second": 1.0452126210233035,
+    "producer_headroom": 2.9538322449839374,
     "sustainable": true,
-    "buffer_seconds": 60.869596174703275,
+    "buffer_seconds": 60.27481751829649,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442550.82617670845,
-    "upload_bytes_per_second": 5802760.875638526,
+    "generation_bytes_per_second": 488598.9201478081,
+    "upload_bytes_per_second": 5862761.4476473695,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.96110524154606,
+    "estimated_full_shards_per_hour": 104.84195426297838,
     "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": 0.9178262779996658,
-      "bytes": 2279800,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.0028891179999846,
       "bytes": 5661960,
@@ -562,6 +556,12 @@
       "bytes": 6506963,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1795985310000106,
+      "bytes": 8204299,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -95,3 +95,4 @@
 {"step": 494500, "loss": 0.5836164355278015, "lr": 0.0003, "grad_norm": 0.18662898146005005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72378885512, "tokens_per_second": 152257.305106254, "data_wait_s": 7.495000318158418e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 494600, "loss": 0.5923532247543335, "lr": 0.0003, "grad_norm": 0.15387534114428109, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72393405116, "tokens_per_second": 151688.3820863191, "data_wait_s": 1.3479999324772507e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 494700, "loss": 0.5760444402694702, "lr": 0.0003, "grad_norm": 0.15010503830561073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72408159018, "tokens_per_second": 153487.12281849774, "data_wait_s": 1.5775998690514825e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 494800, "loss": 0.5754390954971313, "lr": 0.0003, "grad_norm": 0.15269517759605533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72422987702, "tokens_per_second": 154896.79583288563, "data_wait_s": 1.2518999938038178e-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 sha256:4ae4c602cce9212d25836534003622e6a8f82416474bdb5b58084ff2feec0f43
-size 206394
+oid sha256:fa80500e72372ca211e10a845d843a60285caaa16ac654eeeaef7219ae74d7ef
+size 208512