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": 77122.29337503343,
-      "required_pages_per_second_upper_estimate": 0.0053354992040597175
+      "required_token_positions_per_second": 75952.0772054699,
+      "required_pages_per_second_upper_estimate": 0.005254540934173707
     },
     {
       "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": 9640.28667187918,
-      "required_pages_per_second_upper_estimate": 0.004036186717438144
+      "required_token_positions_per_second": 9494.009650683738,
+      "required_pages_per_second_upper_estimate": 0.003974943583275167
     },
     {
       "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": 19280.57334375836,
-      "required_pages_per_second_upper_estimate": 0.004015634964593141
+      "required_token_positions_per_second": 18988.019301367476,
+      "required_pages_per_second_upper_estimate": 0.003954703672732041
     },
     {
       "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": 28920.86001563754,
-      "required_pages_per_second_upper_estimate": 0.0017337613983411501
+      "required_token_positions_per_second": 28482.028952051212,
+      "required_pages_per_second_upper_estimate": 0.0017074541461353865
     },
     {
       "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": 28920.86001563754,
-      "required_pages_per_second_upper_estimate": 0.002248469656796516
+      "required_token_positions_per_second": 28482.028952051212,
+      "required_pages_per_second_upper_estimate": 0.002214352471816536
     },
     {
       "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": 28920.86001563754,
-      "required_pages_per_second_upper_estimate": 0.37096573948047795
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28482.028952051212,
+      "required_pages_per_second_upper_estimate": 0.36533688577687834
     }
   ],
   "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": 0.09709914397683406,
-    "producer_p95_seconds": 0.2707410760012863,
-    "gpu_update_seconds": 0.9559881685762698,
-    "producer_batches_per_second": 10.298751966738042,
-    "gpu_batches_per_second": 1.0460380503341125,
-    "producer_headroom": 9.845485031303157,
+    "producer_batch_seconds": 0.1976480122970372,
+    "producer_p95_seconds": 0.27304690500022843,
+    "gpu_update_seconds": 0.9707173616930423,
+    "producer_batches_per_second": 5.059499401881869,
+    "gpu_batches_per_second": 1.0301659777217598,
+    "producer_headroom": 4.911343910882294,
     "sustainable": true,
-    "buffer_seconds": 60.227254620305,
+    "buffer_seconds": 60.18447642496862,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348954.90151199326,
-    "upload_bytes_per_second": 4905498.855286173,
+    "generation_bytes_per_second": 375611.2547084938,
+    "upload_bytes_per_second": 4914260.644499286,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.87759861011361,
+    "estimated_full_shards_per_hour": 80.59743147793876,
     "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": 1.3044562169998244,
-      "bytes": 6256652,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3504138789976423,
       "bytes": 6609036,
@@ -562,6 +556,12 @@
       "bytes": 11923823,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.1343894960009493,
+      "bytes": 6177604,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 786400, "loss": 0.5295373797416687, "lr": 3e-05, "grad_norm": 0.26525414383700247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115106625793, "tokens_per_second": 151712.22091129285, "data_wait_s": 1.5200999769149348e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786500, "loss": 0.5489715337753296, "lr": 3e-05, "grad_norm": 0.256187664183011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115121285169, "tokens_per_second": 152298.42278352476, "data_wait_s": 1.4785000530537218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786600, "loss": 0.5553094744682312, "lr": 3e-05, "grad_norm": 0.28255089079386364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115135798880, "tokens_per_second": 150367.19231246604, "data_wait_s": 9.563998901285231e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 786700, "loss": 0.5412750840187073, "lr": 3e-05, "grad_norm": 0.27167565713649355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115150412062, "tokens_per_second": 151988.70076290454, "data_wait_s": 1.2495002010837197e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92a741a9f0052e96b5aed11c4c03fd9fb816c718ab399261a3ed7b64a8c655bf
-size 567468
+oid sha256:1993df22c349eb822d5a5b389d3c8b6f80bd5bde57b2e0aeb90f62339635c288
+size 569586