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": 75820.75579855547,
-      "required_pages_per_second_upper_estimate": 0.005245455814535722
+      "required_token_positions_per_second": 75611.26767581806,
+      "required_pages_per_second_upper_estimate": 0.005230962940125335
     },
     {
       "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": 9477.594474819434,
-      "required_pages_per_second_upper_estimate": 0.003968070891928635
+      "required_token_positions_per_second": 9451.408459477258,
+      "required_pages_per_second_upper_estimate": 0.003957107354131047
     },
     {
       "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": 18955.18894963887,
-      "required_pages_per_second_upper_estimate": 0.003947865976261569
+      "required_token_positions_per_second": 18902.816918954515,
+      "required_pages_per_second_upper_estimate": 0.003936958263413346
     },
     {
       "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": 28432.783424458303,
-      "required_pages_per_second_upper_estimate": 0.0017045019519497584
+      "required_token_positions_per_second": 28354.225378431773,
+      "required_pages_per_second_upper_estimate": 0.0016997925170416587
     },
     {
       "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": 28432.783424458303,
-      "required_pages_per_second_upper_estimate": 0.0022105238486543714
+      "required_token_positions_per_second": 28354.225378431773,
+      "required_pages_per_second_upper_estimate": 0.0022044163061161353
     },
     {
       "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": 28432.783424458303,
-      "required_pages_per_second_upper_estimate": 0.3647052170246444
+      "required_token_positions_per_second": 28354.225378431773,
+      "required_pages_per_second_upper_estimate": 0.3636975587592742
     }
   ],
   "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": 0.07142316797649073,
-    "producer_p95_seconds": 0.26087333500254317,
-    "gpu_update_seconds": 0.9723986423438509,
-    "producer_batches_per_second": 14.001059156731253,
-    "gpu_batches_per_second": 1.028384817146206,
-    "producer_headroom": 13.614610915381412,
+    "producer_batch_seconds": 0.09189733157006685,
+    "producer_p95_seconds": 0.2671403390013438,
+    "gpu_update_seconds": 0.9750927641645616,
+    "producer_batches_per_second": 10.881708782126637,
+    "gpu_batches_per_second": 1.025543452634251,
+    "producer_headroom": 10.610675495197649,
     "sustainable": true,
-    "buffer_seconds": 60.288715825318754,
+    "buffer_seconds": 60.45575137820282,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287862.87124864536,
-    "upload_bytes_per_second": 5163337.54164521,
+    "generation_bytes_per_second": 286719.26538184605,
+    "upload_bytes_per_second": 5148755.697139347,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.76867106527825,
+    "estimated_full_shards_per_hour": 61.5232798680452,
     "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": 1.2528174399994896,
-      "bytes": 8102539,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1965843840007437,
       "bytes": 6380308,
@@ -562,6 +556,12 @@
       "bytes": 6822345,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0887722430015856,
+      "bytes": 6046508,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 739600, "loss": 0.5385658740997314, "lr": 5e-05, "grad_norm": 0.22627905395564046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108257564535, "tokens_per_second": 151704.41062751488, "data_wait_s": 1.791400063666515e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 739700, "loss": 0.5473832488059998, "lr": 5e-05, "grad_norm": 0.23200920743857129, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108272288280, "tokens_per_second": 136407.36833084302, "data_wait_s": 1.5714998880866915e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 739800, "loss": 0.5362396240234375, "lr": 5e-05, "grad_norm": 0.23764093765216354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108286777059, "tokens_per_second": 148746.53036725926, "data_wait_s": 1.7947997548617423e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 739900, "loss": 0.5435524582862854, "lr": 5e-05, "grad_norm": 0.233901433441873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108301331256, "tokens_per_second": 149060.28988581957, "data_wait_s": 2.0437997591216117e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b205623e38c9342af86bd8d7d639893aaf2743076166bd633780a1a2a4b2c12
-size 421878
+oid sha256:fc755bd8fd3523ea8da36a86ee2f5ec08149f4d73048280d2651861ba8404dd7
+size 423996