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": 76879.78618595247,
-      "required_pages_per_second_upper_estimate": 0.005318721994024894
+      "required_token_positions_per_second": 76317.81649150292,
+      "required_pages_per_second_upper_estimate": 0.005279843626613534
     },
     {
       "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": 9609.973273244059,
-      "required_pages_per_second_upper_estimate": 0.004023495130445366
+      "required_token_positions_per_second": 9539.727061437865,
+      "required_pages_per_second_upper_estimate": 0.003994084508469822
     },
     {
       "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": 19219.946546488118,
-      "required_pages_per_second_upper_estimate": 0.0040030080015579115
+      "required_token_positions_per_second": 19079.45412287573,
+      "required_pages_per_second_upper_estimate": 0.003973747134753815
     },
     {
       "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": 28829.919819732175,
-      "required_pages_per_second_upper_estimate": 0.0017283096724542626
+      "required_token_positions_per_second": 28619.181184313595,
+      "required_pages_per_second_upper_estimate": 0.0017156762130401833
     },
     {
       "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": 28829.919819732175,
-      "required_pages_per_second_upper_estimate": 0.002241399456568522
+      "required_token_positions_per_second": 28619.181184313595,
+      "required_pages_per_second_upper_estimate": 0.002225015454606022
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28829.919819732175,
-      "required_pages_per_second_upper_estimate": 0.3697992562913787
+      "required_token_positions_per_second": 28619.181184313595,
+      "required_pages_per_second_upper_estimate": 0.3670961273497466
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.44784652406207215,
-    "producer_p95_seconds": 0.3292943729975377,
-    "gpu_update_seconds": 0.959003707706352,
-    "producer_batches_per_second": 2.2329078071875323,
-    "gpu_batches_per_second": 1.042748836072489,
-    "producer_headroom": 2.141366866059304,
+    "producer_batch_seconds": 0.08822450099188472,
+    "producer_p95_seconds": 0.31486475199926645,
+    "gpu_update_seconds": 0.9660653748945862,
+    "producer_batches_per_second": 11.334719819973643,
+    "gpu_batches_per_second": 1.0351266342706016,
+    "producer_headroom": 10.950080352207934,
     "sustainable": true,
-    "buffer_seconds": 60.41723358550018,
+    "buffer_seconds": 60.862118618358934,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373457.8722675481,
+    "generation_bytes_per_second": 373179.6714843427,
     "upload_bytes_per_second": 4933758.640842618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.13536573428948,
+    "estimated_full_shards_per_hour": 80.07567032239639,
     "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": [

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 590600, "loss": 0.5551534295082092, "lr": 0.0003, "grad_norm": 0.16390083626896032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86445968715, "tokens_per_second": 154027.7580806967, "data_wait_s": 8.51499498821795e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 590700, "loss": 0.6077341437339783, "lr": 0.0003, "grad_norm": 0.15848083511875027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86460672697, "tokens_per_second": 153083.06157210996, "data_wait_s": 1.0315998224541545e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 590800, "loss": 0.5838284492492676, "lr": 0.0003, "grad_norm": 0.17198594263842423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86475194571, "tokens_per_second": 150726.01126480653, "data_wait_s": 1.0618001397233456e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 590900, "loss": 0.5800275802612305, "lr": 0.0003, "grad_norm": 0.18188725123569327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86489847229, "tokens_per_second": 152318.04262182084, "data_wait_s": 7.337002898566425e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:753aaf872e0f370887d51f0e2c4d0b8d098a173f1d986776585c7b2773602d44
-size 719412
+oid sha256:e5932eb4459f4dea02b1ef4a2589b583dbef5e9b395ae3c9024b804dfe98b0dc
+size 721530