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": 77481.38831978457,
-      "required_pages_per_second_upper_estimate": 0.005360342225552668
+      "required_token_positions_per_second": 76898.19041832078,
+      "required_pages_per_second_upper_estimate": 0.005319995241523836
     },
     {
       "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": 9685.173539973071,
-      "required_pages_per_second_upper_estimate": 0.004054979911764662
+      "required_token_positions_per_second": 9612.273802290098,
+      "required_pages_per_second_upper_estimate": 0.004024458313916433
     },
     {
       "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": 19370.347079946143,
-      "required_pages_per_second_upper_estimate": 0.004034332466348427
+      "required_token_positions_per_second": 19224.547604580195,
+      "required_pages_per_second_upper_estimate": 0.004003966280620429
     },
     {
       "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": 29055.520619919218,
-      "required_pages_per_second_upper_estimate": 0.0017418340959529995
+      "required_token_positions_per_second": 28836.821406870295,
+      "required_pages_per_second_upper_estimate": 0.0017287234120650795
     },
     {
       "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": 29055.520619919218,
-      "required_pages_per_second_upper_estimate": 0.002258938926470015
+      "required_token_positions_per_second": 28836.821406870295,
+      "required_pages_per_second_upper_estimate": 0.002241936024611637
     },
     {
       "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": 29055.520619919218,
-      "required_pages_per_second_upper_estimate": 0.3726930211249114
+      "required_token_positions_per_second": 28836.821406870295,
+      "required_pages_per_second_upper_estimate": 0.3698877824408396
     }
   ],
   "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": 0.11775623846079952,
-    "producer_p95_seconds": 0.2482576299989887,
-    "gpu_update_seconds": 0.9515575494815164,
-    "producer_batches_per_second": 8.492119084908568,
-    "gpu_batches_per_second": 1.0509085872366615,
-    "producer_headroom": 8.080740026340814,
+    "producer_batch_seconds": 0.36577008649217646,
+    "producer_p95_seconds": 0.2444531779983663,
+    "gpu_update_seconds": 0.9587741869987425,
+    "producer_batches_per_second": 2.733957852021858,
+    "gpu_batches_per_second": 1.0429984594498805,
+    "producer_headroom": 2.6212482168610856,
     "sustainable": true,
-    "buffer_seconds": 60.89968316681705,
+    "buffer_seconds": 60.402773780920775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311396.8137551801,
-    "upload_bytes_per_second": 5504119.239878215,
+    "generation_bytes_per_second": 303633.8552455705,
+    "upload_bytes_per_second": 5552280.816436256,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.81850728503754,
+    "estimated_full_shards_per_hour": 65.15275710130058,
     "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": 1.260699045000365,
-      "bytes": 2670941,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.4827766799971869,
       "bytes": 15946512,
@@ -562,6 +556,12 @@
       "bytes": 2665696,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.1741747699998086,
+      "bytes": 5919198,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -254,3 +254,4 @@
 {"step": 582500, "loss": 0.6072389483451843, "lr": 0.0003, "grad_norm": 0.1896262142150918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85261442459, "tokens_per_second": 148580.8885633686, "data_wait_s": 8.775001333560795e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 582600, "loss": 0.542369544506073, "lr": 0.0003, "grad_norm": 0.15535529138038562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85276068054, "tokens_per_second": 152024.50757107403, "data_wait_s": 9.142000635620207e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 582700, "loss": 0.5577417612075806, "lr": 0.0003, "grad_norm": 0.17495125828870997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85290500663, "tokens_per_second": 150200.08823705386, "data_wait_s": 1.86079996638e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 582800, "loss": 0.6232706308364868, "lr": 0.0003, "grad_norm": 0.17642637879539783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85305100031, "tokens_per_second": 152282.8505410679, "data_wait_s": 1.42680000863038e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6351d317143960ab686b6ddf255e9d8810b111428018528ada10aec5b30e630
-size 546288
+oid sha256:1a3ef9b50fe7cf0511ca54e1f6c8a7d6cc8bb5c61a7812f699da06b081b63d98
+size 548406