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": 75939.4682566325,
-      "required_pages_per_second_upper_estimate": 0.005253668617836344
+      "required_token_positions_per_second": 76104.76119769279,
+      "required_pages_per_second_upper_estimate": 0.005265103967031358
     },
     {
       "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": 9492.433532079063,
-      "required_pages_per_second_upper_estimate": 0.003974283695328489
+      "required_token_positions_per_second": 9513.095149711598,
+      "required_pages_per_second_upper_estimate": 0.003982934283167591
     },
     {
       "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": 18984.867064158127,
-      "required_pages_per_second_upper_estimate": 0.003954047144851419
+      "required_token_positions_per_second": 19026.190299423197,
+      "required_pages_per_second_upper_estimate": 0.003962653684990136
     },
     {
       "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": 28477.30059623719,
-      "required_pages_per_second_upper_estimate": 0.001707170688424117
+      "required_token_positions_per_second": 28539.285449134797,
+      "required_pages_per_second_upper_estimate": 0.0017108865857099385
     },
     {
       "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": 28477.30059623719,
-      "required_pages_per_second_upper_estimate": 0.0022139848629498353
+      "required_token_positions_per_second": 28539.285449134797,
+      "required_pages_per_second_upper_estimate": 0.00221880391262007
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28477.30059623719,
-      "required_pages_per_second_upper_estimate": 0.365276235505409
+      "required_token_positions_per_second": 28539.285449134797,
+      "required_pages_per_second_upper_estimate": 0.36607131064422976
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06109552251569994,
-    "producer_p95_seconds": 0.21143909599777544,
-    "gpu_update_seconds": 0.9708785390864341,
-    "producer_batches_per_second": 16.367811564964132,
-    "gpu_batches_per_second": 1.0299949579078844,
-    "producer_headroom": 15.891156980234417,
+    "producer_batch_seconds": 0.10997083356247117,
+    "producer_p95_seconds": 0.24917777500013472,
+    "gpu_update_seconds": 0.9687698750999452,
+    "producer_batches_per_second": 9.093320179590433,
+    "gpu_batches_per_second": 1.0322368869044705,
+    "producer_headroom": 8.809334654625635,
     "sustainable": true,
-    "buffer_seconds": 60.19446942335891,
+    "buffer_seconds": 60.0637322561966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422181.7553335597,
-    "upload_bytes_per_second": 3943382.5201663566,
+    "generation_bytes_per_second": 420243.4869565159,
+    "upload_bytes_per_second": 3841996.125011375,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.59037680630773,
+    "estimated_full_shards_per_hour": 90.17446953317267,
     "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": 1.8264212399990356,
-      "bytes": 16670948,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.714815281000483,
       "bytes": 7600635,
@@ -562,6 +556,12 @@
       "bytes": 5917528,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.7868269310019969,
+      "bytes": 5421842,
+      "pages": 40,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 815400, "loss": 0.5271826982498169, "lr": 3e-05, "grad_norm": 0.27651901645506544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119351405307, "tokens_per_second": 151689.44992152092, "data_wait_s": 1.5000998246250674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 815500, "loss": 0.5773677825927734, "lr": 3e-05, "grad_norm": 0.2918577702769365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119366169299, "tokens_per_second": 152090.05018573714, "data_wait_s": 1.5142999473027885e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 815600, "loss": 0.5469526648521423, "lr": 3e-05, "grad_norm": 0.28253347800641926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119380869842, "tokens_per_second": 152208.78769467058, "data_wait_s": 2.128100095433183e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 815700, "loss": 0.5718827247619629, "lr": 3e-05, "grad_norm": 0.2807549818283929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119395443058, "tokens_per_second": 151021.98915335568, "data_wait_s": 1.8255999748362228e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:979b9d7f2d63d18983aae382a6f088654197a45168ae1a8f65ae540504e36e5c
-size 332922
+oid sha256:4c638b6e056055dd565c5a3b009263e9e23eeb895608633a391a5ae3a6a0d695
+size 335040