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": 76267.79219567539,
-      "required_pages_per_second_upper_estimate": 0.005276382829755827
+      "required_token_positions_per_second": 75651.63531644401,
+      "required_pages_per_second_upper_estimate": 0.005233755667169669
     },
     {
       "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": 9533.474024459423,
-      "required_pages_per_second_upper_estimate": 0.003991466492465209
+      "required_token_positions_per_second": 9456.454414555501,
+      "required_pages_per_second_upper_estimate": 0.003959219990150783
     },
     {
       "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": 19066.948048918846,
-      "required_pages_per_second_upper_estimate": 0.003971142449355944
+      "required_token_positions_per_second": 18912.908829111002,
+      "required_pages_per_second_upper_estimate": 0.003939060142157374
     },
     {
       "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": 28600.422073378268,
-      "required_pages_per_second_upper_estimate": 0.0017145516329831144
+      "required_token_positions_per_second": 28369.363243666503,
+      "required_pages_per_second_upper_estimate": 0.001700700008948299
     },
     {
       "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": 28600.422073378268,
-      "required_pages_per_second_upper_estimate": 0.0022235570162434095
+      "required_token_positions_per_second": 28369.363243666503,
+      "required_pages_per_second_upper_estimate": 0.0022055932085537034
     },
     {
       "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": 28600.422073378268,
-      "required_pages_per_second_upper_estimate": 0.36685550561663227
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28369.363243666503,
+      "required_pages_per_second_upper_estimate": 0.36389173104073197
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07252109690631414,
-    "producer_p95_seconds": 0.21685623299708823,
-    "gpu_update_seconds": 0.9666990203524025,
-    "producer_batches_per_second": 13.789090935729265,
-    "gpu_batches_per_second": 1.0344481363345728,
-    "producer_headroom": 13.329900699119674,
+    "producer_batch_seconds": 0.4399874340547285,
+    "producer_p95_seconds": 0.23579259799953434,
+    "gpu_update_seconds": 0.9745724555933575,
+    "producer_batches_per_second": 2.2727921813231,
+    "gpu_batches_per_second": 1.026090973801595,
+    "producer_headroom": 2.215000657205437,
     "sustainable": true,
-    "buffer_seconds": 60.90203828220136,
+    "buffer_seconds": 60.423492246788165,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276305.500825575,
-    "upload_bytes_per_second": 4930656.622660016,
+    "generation_bytes_per_second": 276776.31733265804,
+    "upload_bytes_per_second": 4943741.715317381,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.28872841430128,
+    "estimated_full_shards_per_hour": 59.389754676673945,
     "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": 1.550046686999849,
-      "bytes": 10810645,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.2842068059999292,
       "bytes": 7447718,
@@ -562,6 +556,12 @@
       "bytes": 6041473,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3613106869997864,
+      "bytes": 11006649,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -165,3 +165,4 @@
 {"step": 736500, "loss": 0.5354722738265991, "lr": 5e-05, "grad_norm": 0.22226708118344876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107803168898, "tokens_per_second": 151647.89899230332, "data_wait_s": 1.8583999917609617e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736600, "loss": 0.4877394735813141, "lr": 5e-05, "grad_norm": 0.2158283437587591, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107817819827, "tokens_per_second": 150850.3332295701, "data_wait_s": 1.6772999515524134e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736700, "loss": 0.5290622115135193, "lr": 5e-05, "grad_norm": 0.2205401782882761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107832514860, "tokens_per_second": 151377.62170462927, "data_wait_s": 1.65250021382235e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 736800, "loss": 0.5798033475875854, "lr": 5e-05, "grad_norm": 0.22902320557757003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107847215707, "tokens_per_second": 150690.55532955113, "data_wait_s": 1.414399957866408e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3ebf14e86bbda0f113ba801b8fd683e3d943e0e00731cfc094dbf88066f769e
-size 356220
+oid sha256:678d5358e5a97a270b35ae5d718d0a00ef65a8befe2e3161c5c46aea46e0d2ff
+size 358338