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": 77134.5857151549,
-      "required_pages_per_second_upper_estimate": 0.005336349616671479
+      "required_token_positions_per_second": 75090.18750747145,
+      "required_pages_per_second_upper_estimate": 0.005194913405006548
     },
     {
       "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": 9641.823214394362,
-      "required_pages_per_second_upper_estimate": 0.0040368300357025945
+      "required_token_positions_per_second": 9386.273438433931,
+      "required_pages_per_second_upper_estimate": 0.003929836680994116
     },
     {
       "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": 19283.646428788725,
-      "required_pages_per_second_upper_estimate": 0.004016275007162264
+      "required_token_positions_per_second": 18772.546876867862,
+      "required_pages_per_second_upper_estimate": 0.0039098264490987295
     },
     {
       "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": 28925.469643183085,
-      "required_pages_per_second_upper_estimate": 0.0017340377384740119
+      "required_token_positions_per_second": 28158.82031530179,
+      "required_pages_per_second_upper_estimate": 0.0016880782818732717
     },
     {
       "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": 28925.469643183085,
-      "required_pages_per_second_upper_estimate": 0.002248828035062573
+      "required_token_positions_per_second": 28158.82031530179,
+      "required_pages_per_second_upper_estimate": 0.002189224421953818
     },
     {
       "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": 28925.469643183085,
-      "required_pages_per_second_upper_estimate": 0.37102486683319974
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28158.82031530179,
+      "required_pages_per_second_upper_estimate": 0.361191112419053
     }
   ],
   "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": 0.10176557435909217,
-    "producer_p95_seconds": 0.37464002900378546,
-    "gpu_update_seconds": 0.9558358201632813,
-    "producer_batches_per_second": 9.826505734359428,
-    "gpu_batches_per_second": 1.0462047758674438,
-    "producer_headroom": 9.39252616794063,
+    "producer_batch_seconds": 0.2112046502890621,
+    "producer_p95_seconds": 0.3523182260032627,
+    "gpu_update_seconds": 0.9818593140770103,
+    "producer_batches_per_second": 4.734744233289205,
+    "gpu_batches_per_second": 1.018475850524515,
+    "producer_headroom": 4.648852725227419,
     "sustainable": true,
-    "buffer_seconds": 60.21765667028672,
+    "buffer_seconds": 60.87527747277464,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 641124.7491327933,
-    "upload_bytes_per_second": 5029416.853268864,
+    "generation_bytes_per_second": 630518.8578147442,
+    "upload_bytes_per_second": 5079089.00197146,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.78522327178882,
+    "estimated_full_shards_per_hour": 67.64733457961914,
     "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": 1.1930818900000304,
-      "bytes": 5619417,
-      "pages": 14,
-      "shards": 2
-    },
     {
       "seconds": 1.3075707610041718,
       "bytes": 7296914,
@@ -562,6 +556,12 @@
       "bytes": 7827828,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.3130354629975045,
+      "bytes": 10468764,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -427,3 +427,4 @@
 {"step": 967700, "loss": 0.507817804813385, "lr": 3e-05, "grad_norm": 0.34821866907319726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141645206040, "tokens_per_second": 151648.98401438427, "data_wait_s": 1.2882002920378e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 967800, "loss": 0.5485464930534363, "lr": 3e-05, "grad_norm": 0.3445373450443752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141659705637, "tokens_per_second": 150332.2215349967, "data_wait_s": 1.1493997590150684e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 967900, "loss": 0.51126629114151, "lr": 3e-05, "grad_norm": 0.3486149697053852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141674228114, "tokens_per_second": 149765.28158579825, "data_wait_s": 1.2918004358652979e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 968000, "loss": 0.5415030121803284, "lr": 3e-05, "grad_norm": 0.33291041638664554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141688819063, "tokens_per_second": 151903.86568822814, "data_wait_s": 9.25700442166999e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b5931d2fe94dc32344a79a45b4ab84a675f35109ab29915779c28c9c6a958c2
-size 915051
+oid sha256:00d650c9d5b707e93709d72cbb343a785e79d05d6a1fcdf7da757f094591f5ae
+size 917169