Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74738.12227497603,
-      "required_pages_per_second_upper_estimate": 0.006478111016755797
+      "required_token_positions_per_second": 74572.3251876952,
+      "required_pages_per_second_upper_estimate": 0.00646374014543917
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9342.265284372004,
-      "required_pages_per_second_upper_estimate": 0.0037267693012494035
+      "required_token_positions_per_second": 9321.5406484619,
+      "required_pages_per_second_upper_estimate": 0.0037185019341239426
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18684.53056874401,
-      "required_pages_per_second_upper_estimate": 0.0043223373105720775
+      "required_token_positions_per_second": 18643.0812969238,
+      "required_pages_per_second_upper_estimate": 0.004312748751018742
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28026.79585311601,
-      "required_pages_per_second_upper_estimate": 0.0017066848222211783
+      "required_token_positions_per_second": 27964.6219453857,
+      "required_pages_per_second_upper_estimate": 0.001702898757441686
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28026.79585311601,
-      "required_pages_per_second_upper_estimate": 0.0021575803279093335
+      "required_token_positions_per_second": 27964.6219453857,
+      "required_pages_per_second_upper_estimate": 0.002152794008383862
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28026.79585311601,
-      "required_pages_per_second_upper_estimate": 0.3479990048439352
+      "required_token_positions_per_second": 27964.6219453857,
+      "required_pages_per_second_upper_estimate": 0.3472270129926084
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1555222331638788,
-    "producer_p95_seconds": 0.43609423100133426,
-    "gpu_update_seconds": 0.9864845109265711,
-    "producer_batches_per_second": 6.429948822470082,
-    "gpu_batches_per_second": 1.013700660196615,
-    "producer_headroom": 6.34304491941728,
+    "producer_batch_seconds": 0.10193264426601445,
+    "producer_p95_seconds": 0.4177872920045047,
+    "gpu_update_seconds": 0.9886777677165078,
+    "producer_batches_per_second": 9.810399869449986,
+    "gpu_batches_per_second": 1.011451893279286,
+    "producer_headroom": 9.699324243334132,
     "sustainable": true,
-    "buffer_seconds": 60.17555516652084,
+    "buffer_seconds": 60.30934383070698,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337392.52271577355,
-    "upload_bytes_per_second": 5363678.822132393,
+    "generation_bytes_per_second": 337248.03770769,
+    "upload_bytes_per_second": 5287585.875586291,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.39658127884775,
+    "estimated_full_shards_per_hour": 72.36557815955186,
     "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": 0.8739250739999989,
-      "bytes": 5944089,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 0.9463060489979398,
       "bytes": 743440,
@@ -562,6 +556,12 @@
       "bytes": 7653463,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0381377000012435,
+      "bytes": 839375,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 354600, "loss": 0.5678000226616859, "lr": 0.0003, "grad_norm": 0.1558815531563137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51903795748, "tokens_per_second": 149932.9615391219, "data_wait_s": 1.1664997145999223e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354700, "loss": 0.56468465924263, "lr": 0.0003, "grad_norm": 0.16401846710138815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51918300294, "tokens_per_second": 146860.8932387149, "data_wait_s": 1.1041003745049238e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354800, "loss": 0.5827666632831097, "lr": 0.0003, "grad_norm": 0.14574788928946014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51932766104, "tokens_per_second": 146994.49737078985, "data_wait_s": 9.767994924914092e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 354900, "loss": 0.584769681096077, "lr": 0.0003, "grad_norm": 0.1662222847641233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51947299203, "tokens_per_second": 140361.33777305816, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47f3ea81f1e613b776f19ba6fc289c5047bb02ec71ba7bc59582550672691d12
-size 848610
+oid sha256:afe78cf82c7e318d10c88cd897bdb9a5ed277181e698a6f1543234aee1b00591
+size 850728