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": 74799.10180867862,
-      "required_pages_per_second_upper_estimate": 0.006483396568185923
+      "required_token_positions_per_second": 72106.31265299165,
+      "required_pages_per_second_upper_estimate": 0.006249992428982695
     },
     {
       "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": 9349.887726084828,
-      "required_pages_per_second_upper_estimate": 0.003729810007214308
+      "required_token_positions_per_second": 9013.289081623956,
+      "required_pages_per_second_upper_estimate": 0.003595535775340656
     },
     {
       "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": 18699.775452169655,
-      "required_pages_per_second_upper_estimate": 0.004325863946051817
+      "required_token_positions_per_second": 18026.578163247912,
+      "required_pages_per_second_upper_estimate": 0.004170131601127401
     },
     {
       "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": 28049.663178254483,
-      "required_pages_per_second_upper_estimate": 0.0017080773223465283
+      "required_token_positions_per_second": 27039.867244871868,
+      "required_pages_per_second_upper_estimate": 0.0016465860479933394
     },
     {
       "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": 28049.663178254483,
-      "required_pages_per_second_upper_estimate": 0.0021593407178993004
+      "required_token_positions_per_second": 27039.867244871868,
+      "required_pages_per_second_upper_estimate": 0.0020816038316534566
     },
     {
       "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": 28049.663178254483,
-      "required_pages_per_second_upper_estimate": 0.34828294049014097
+      "required_token_positions_per_second": 27039.867244871868,
+      "required_pages_per_second_upper_estimate": 0.3357446545671166
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 59,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08941659166401905,
-    "producer_p95_seconds": 0.24721415499880095,
-    "gpu_update_seconds": 0.9856802851534463,
-    "producer_batches_per_second": 11.18360677129675,
-    "gpu_batches_per_second": 1.0145277480560793,
-    "producer_headroom": 11.023460711375794,
+    "producer_batch_seconds": 0.36509881958590995,
+    "producer_p95_seconds": 0.31781539599978714,
+    "gpu_update_seconds": 1.0224902271013725,
+    "producer_batches_per_second": 2.738984478597291,
+    "gpu_batches_per_second": 0.9780044576414882,
+    "producer_headroom": 2.8005848615480784,
     "sustainable": true,
-    "buffer_seconds": 60.12649739436022,
+    "buffer_seconds": 60.32692339898098,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351067.46730236325,
-    "upload_bytes_per_second": 5846551.247474989,
+    "generation_bytes_per_second": 355498.6706685716,
+    "upload_bytes_per_second": 5775043.062551655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.33090605071233,
+    "estimated_full_shards_per_hour": 76.28173913996564,
     "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": 1.2563196339997376,
-      "bytes": 7398482,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.2963822889996663,
       "bytes": 6635646,
@@ -562,6 +556,12 @@
       "bytes": 11937829,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2463495720003266,
+      "bytes": 2130021,
+      "pages": 22,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -203,3 +203,4 @@
 {"step": 375300, "loss": 0.5982198733836412, "lr": 0.0003, "grad_norm": 0.16705821078006447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54932780320, "tokens_per_second": 143978.49520058522, "data_wait_s": 7.730999641353264e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 375400, "loss": 0.5743808448314667, "lr": 0.0003, "grad_norm": 0.15233685722282306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54947279592, "tokens_per_second": 147841.1089620428, "data_wait_s": 9.740000677993521e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 375500, "loss": 0.5590781159698963, "lr": 0.0003, "grad_norm": 0.15987953919798303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54961898171, "tokens_per_second": 148240.55917270106, "data_wait_s": 1.4318000467028469e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 375600, "loss": 0.5565471760928631, "lr": 0.0003, "grad_norm": 0.15656504648731412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54976547112, "tokens_per_second": 148730.40929264852, "data_wait_s": 1.4162997103994712e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae4e53cf0efcc999f6f496509397e8f8a7a3dbfdc6de877dc2c8e1a28eaa6804
-size 437487
+oid sha256:174279524ee3202a8c007af816aafa946b84c7efe54001549e45a026a54bda04
+size 439605