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": 77309.27550044844,
-      "required_pages_per_second_upper_estimate": 0.005348435061354233
+      "required_token_positions_per_second": 75791.51928327057,
+      "required_pages_per_second_upper_estimate": 0.005243433164570251
     },
     {
       "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": 9663.659437556054,
-      "required_pages_per_second_upper_estimate": 0.0040459724063482035
+      "required_token_positions_per_second": 9473.93991040882,
+      "required_pages_per_second_upper_estimate": 0.003966540802125894
     },
     {
       "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": 19327.31887511211,
-      "required_pages_per_second_upper_estimate": 0.004025370826011566
+      "required_token_positions_per_second": 18947.87982081764,
+      "required_pages_per_second_upper_estimate": 0.003946343677482797
     },
     {
       "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": 28990.97831266816,
-      "required_pages_per_second_upper_estimate": 0.0017379648831836957
+      "required_token_positions_per_second": 28421.81973122646,
+      "required_pages_per_second_upper_estimate": 0.0017038446952810996
     },
     {
       "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": 28990.97831266816,
-      "required_pages_per_second_upper_estimate": 0.0022539210459728526
+      "required_token_positions_per_second": 28421.81973122646,
+      "required_pages_per_second_upper_estimate": 0.0022096714697297857
     },
     {
       "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": 28990.97831266816,
-      "required_pages_per_second_upper_estimate": 0.37186514170762514
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28421.81973122646,
+      "required_pages_per_second_upper_estimate": 0.364564586539763
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 720750,
-    "producer_batch_seconds": 0.05302272368743388,
-    "producer_p95_seconds": 0.1645179819997793,
-    "gpu_update_seconds": 0.9536759919522508,
-    "producer_batches_per_second": 18.859838394854,
-    "gpu_batches_per_second": 1.048574157720926,
-    "producer_headroom": 17.98617508927153,
+    "producer_batch_seconds": 0.4758416528202929,
+    "producer_p95_seconds": 0.1706308729999364,
+    "gpu_update_seconds": 0.9727737443082759,
+    "producer_batches_per_second": 2.1015394387461526,
+    "gpu_batches_per_second": 1.0279882715287347,
+    "producer_headroom": 2.0443223886406074,
     "sustainable": true,
-    "buffer_seconds": 60.0815874929918,
+    "buffer_seconds": 60.3119721471131,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433303.45547621214,
-    "upload_bytes_per_second": 3085286.3142732033,
+    "generation_bytes_per_second": 376954.87482256984,
+    "upload_bytes_per_second": 3597951.1739881802,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.97683475699209,
+    "estimated_full_shards_per_hour": 80.88574107654401,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 1119612,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 2.561698224999418,
+      "bytes": 16238211,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 929800, "loss": 0.49387720227241516, "lr": 3e-05, "grad_norm": 0.3172120206524664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136096317673, "tokens_per_second": 151925.53029914363, "data_wait_s": 1.4021999959368259e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 929900, "loss": 0.5518815517425537, "lr": 3e-05, "grad_norm": 0.3177301556255189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136110929325, "tokens_per_second": 152065.37960935698, "data_wait_s": 1.3428000784188043e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 930000, "loss": 0.5029640793800354, "lr": 3e-05, "grad_norm": 0.33582552487306594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136125694727, "tokens_per_second": 152754.54049881033, "data_wait_s": 1.1747000826289877e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 930100, "loss": 0.5324306488037109, "lr": 3e-05, "grad_norm": 0.33269737730314974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136140482529, "tokens_per_second": 134912.15963558116, "data_wait_s": 1.0929000382020604e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d1e622c6ee9c3d14922d8b620e760070f98042ddec82c0e684e9f7f211a048e
-size 106848
+oid sha256:3b426fe38dad94c7d9b07dd049304da0073c6acae1075c38eb3aaa2455a9894d
+size 108966