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": 74726.03048801875,
-      "required_pages_per_second_upper_estimate": 0.005169720177440112
+      "required_token_positions_per_second": 76897.31413435678,
+      "required_pages_per_second_upper_estimate": 0.005319934618165948
     },
     {
       "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": 9340.753811002343,
-      "required_pages_per_second_upper_estimate": 0.003910778563546423
+      "required_token_positions_per_second": 9612.164266794598,
+      "required_pages_per_second_upper_estimate": 0.004024412453691828
     },
     {
       "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": 18681.507622004687,
-      "required_pages_per_second_upper_estimate": 0.003890865373177332
+      "required_token_positions_per_second": 19224.328533589196,
+      "required_pages_per_second_upper_estimate": 0.004003920653910295
     },
     {
       "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": 28022.261433007032,
-      "required_pages_per_second_upper_estimate": 0.001679891785393041
+      "required_token_positions_per_second": 28836.492800383796,
+      "required_pages_per_second_upper_estimate": 0.001728703712607967
     },
     {
       "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": 28022.261433007032,
-      "required_pages_per_second_upper_estimate": 0.002178607569514444
+      "required_token_positions_per_second": 28836.492800383796,
+      "required_pages_per_second_upper_estimate": 0.002241910476902699
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28022.261433007032,
-      "required_pages_per_second_upper_estimate": 0.35943948170247986
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28836.492800383796,
+      "required_pages_per_second_upper_estimate": 0.36988356742966094
     }
   ],
   "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": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08847235714861768,
-    "producer_p95_seconds": 0.3253873880021274,
-    "gpu_update_seconds": 0.986644138842906,
-    "producer_batches_per_second": 11.302965493732462,
-    "gpu_batches_per_second": 1.013536654839664,
-    "producer_headroom": 11.152004655934746,
+    "producer_batch_seconds": 0.10985123967211052,
+    "producer_p95_seconds": 0.32642102200043155,
+    "gpu_update_seconds": 0.9587851127177305,
+    "producer_batches_per_second": 9.103219981721189,
+    "gpu_batches_per_second": 1.0429865740879556,
+    "producer_headroom": 8.728031796268848,
     "sustainable": true,
-    "buffer_seconds": 60.18529246941727,
+    "buffer_seconds": 60.40346210121702,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276725.2012079336,
-    "upload_bytes_per_second": 5478309.788456783,
+    "generation_bytes_per_second": 277622.73448463005,
+    "upload_bytes_per_second": 5504956.273912091,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.37878634623056,
+    "estimated_full_shards_per_hour": 59.57137609390427,
     "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": 1.3021024750014476,
-      "bytes": 5911875,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0596985519987356,
       "bytes": 6289450,
@@ -562,6 +556,12 @@
       "bytes": 1060442,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.0037741340020148,
+      "bytes": 6334013,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -323,3 +323,4 @@
 {"step": 472500, "loss": 0.5736663341522217, "lr": 0.0003, "grad_norm": 0.16556079692463208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69159366715, "tokens_per_second": 153450.70044017528, "data_wait_s": 1.2171996786491945e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 472600, "loss": 0.5829270482063293, "lr": 0.0003, "grad_norm": 0.15995475005757198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69174063494, "tokens_per_second": 152817.12225836012, "data_wait_s": 1.0965999535983428e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 472700, "loss": 0.5891632437705994, "lr": 0.0003, "grad_norm": 0.1480681987207423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69188784306, "tokens_per_second": 152529.02592809053, "data_wait_s": 1.207699824590236e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 472800, "loss": 0.5954067707061768, "lr": 0.0003, "grad_norm": 0.16198262846134612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69203464134, "tokens_per_second": 152406.81980625115, "data_wait_s": 1.3474000297719613e-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:a190f4bcb06e5aa44952595f0a09c58e16cc6f655dd72da4e939868119f92b7c
-size 693213
+oid sha256:86a630e92a107051b26d13862254635dacb8a023339ccdae3a09ffb385d624b9
+size 695331