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": 75628.38869600375,
-      "required_pages_per_second_upper_estimate": 0.006555277053770864
+      "required_token_positions_per_second": 74946.10716401812,
+      "required_pages_per_second_upper_estimate": 0.006496138619805083
     },
     {
       "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": 9453.548587000469,
-      "required_pages_per_second_upper_estimate": 0.003771161874501543
+      "required_token_positions_per_second": 9368.263395502265,
+      "required_pages_per_second_upper_estimate": 0.0037371403364856648
     },
     {
       "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": 18907.097174000937,
-      "required_pages_per_second_upper_estimate": 0.004373824177659799
+      "required_token_positions_per_second": 18736.52679100453,
+      "required_pages_per_second_upper_estimate": 0.004334365721489792
     },
     {
       "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": 28360.645761001408,
-      "required_pages_per_second_upper_estimate": 0.0017270145300363011
+      "required_token_positions_per_second": 28104.790186506794,
+      "required_pages_per_second_upper_estimate": 0.0017114342679200346
     },
     {
       "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": 28360.645761001408,
-      "required_pages_per_second_upper_estimate": 0.002183281017974045
+      "required_token_positions_per_second": 28104.790186506794,
+      "required_pages_per_second_upper_estimate": 0.0021635845475958893
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28360.645761001408,
-      "required_pages_per_second_upper_estimate": 0.35214430337610547
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28104.790186506794,
+      "required_pages_per_second_upper_estimate": 0.34896743343440645
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06744973229683637,
-    "producer_p95_seconds": 0.22565372499957448,
-    "gpu_update_seconds": 0.9748720192407831,
-    "producer_batches_per_second": 14.825855729110188,
-    "gpu_batches_per_second": 1.0257756713325161,
-    "producer_headroom": 14.453311911610182,
+    "producer_batch_seconds": 0.3555096230936101,
+    "producer_p95_seconds": 0.2918402870018326,
+    "gpu_update_seconds": 0.9837468921320714,
+    "producer_batches_per_second": 2.812863379894185,
+    "gpu_batches_per_second": 1.0165216357966866,
+    "producer_headroom": 2.767145607963019,
     "sustainable": true,
-    "buffer_seconds": 60.44206519292855,
+    "buffer_seconds": 60.008560420056355,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342753.1245784973,
-    "upload_bytes_per_second": 5254260.278932557,
+    "generation_bytes_per_second": 319596.3544944696,
+    "upload_bytes_per_second": 5211964.39577559,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.54684165016356,
+    "estimated_full_shards_per_hour": 68.57793785214963,
     "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": 0.9634915199985699,
-      "bytes": 7597611,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.761399758999687,
       "bytes": 1061264,
@@ -562,6 +556,12 @@
       "bytes": 971215,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.235119819000829,
+      "bytes": 5680772,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -163,3 +163,4 @@
 {"step": 211300, "loss": 0.6035649683326483, "lr": 0.0003, "grad_norm": 0.1671182665905871, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30927287802, "tokens_per_second": 150791.66435062894, "data_wait_s": 1.540799712529406e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 211400, "loss": 0.5888801328837872, "lr": 0.0003, "grad_norm": 0.16051327812007865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30941897769, "tokens_per_second": 148663.0822992475, "data_wait_s": 9.882998710963875e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 211500, "loss": 0.6214869990944862, "lr": 0.0003, "grad_norm": 0.1705022461807806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30956467534, "tokens_per_second": 147575.71681998717, "data_wait_s": 1.6814999980852008e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 211600, "loss": 0.5997838526964188, "lr": 0.0003, "grad_norm": 0.14328802200601443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30971026863, "tokens_per_second": 147683.235183447, "data_wait_s": 1.5331999748013914e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c317d158786592b6624133e43329f7562bb21581cf062883c496f7e945bc6e97
-size 351984
+oid sha256:2929b06506f7fa43237b5493117d38bfe558c99e9897576afd779360476815aa
+size 354102