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": 75514.87996795733,
-      "required_pages_per_second_upper_estimate": 0.006545438404909041
+      "required_token_positions_per_second": 75257.5272233036,
+      "required_pages_per_second_upper_estimate": 0.0065231317212570234
     },
     {
       "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": 9439.359995994666,
-      "required_pages_per_second_upper_estimate": 0.0037655018334109883
+      "required_token_positions_per_second": 9407.19090291295,
+      "required_pages_per_second_upper_estimate": 0.0037526691012098887
     },
     {
       "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": 18878.719991989332,
-      "required_pages_per_second_upper_estimate": 0.004367259616022761
+      "required_token_positions_per_second": 18814.3818058259,
+      "required_pages_per_second_upper_estimate": 0.0043523761089672534
     },
     {
       "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": 28318.079987983998,
-      "required_pages_per_second_upper_estimate": 0.0017244224977848923
+      "required_token_positions_per_second": 28221.572708738848,
+      "required_pages_per_second_upper_estimate": 0.0017185457108134252
     },
     {
       "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": 28318.079987983998,
-      "required_pages_per_second_upper_estimate": 0.0021800041869375637
+      "required_token_positions_per_second": 28221.572708738848,
+      "required_pages_per_second_upper_estimate": 0.002172574789432025
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28318.079987983998,
-      "required_pages_per_second_upper_estimate": 0.35161577893370743
+      "required_token_positions_per_second": 28221.572708738848,
+      "required_pages_per_second_upper_estimate": 0.35041748151456903
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12827800925779798,
-    "producer_p95_seconds": 0.32516657099768054,
-    "gpu_update_seconds": 0.9763373792196248,
-    "producer_batches_per_second": 7.795568435976569,
-    "gpu_batches_per_second": 1.0242361106765046,
-    "producer_headroom": 7.611104856308593,
+    "producer_batch_seconds": 0.2680376398827775,
+    "producer_p95_seconds": 0.35673518600015086,
+    "gpu_update_seconds": 0.9796760898246737,
+    "producer_batches_per_second": 3.730819299995836,
+    "gpu_batches_per_second": 1.0207455406806587,
+    "producer_headroom": 3.654994463662347,
     "sustainable": true,
-    "buffer_seconds": 60.53291751161674,
+    "buffer_seconds": 60.739917569129766,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379891.0235785658,
-    "upload_bytes_per_second": 4921676.045441605,
+    "generation_bytes_per_second": 377130.5693988787,
+    "upload_bytes_per_second": 4819751.596926739,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.51577024953585,
+    "estimated_full_shards_per_hour": 80.92344104265949,
     "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": 1.37565966299735,
-      "bytes": 13270818,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.7815917669977352,
       "bytes": 913950,
@@ -562,6 +556,12 @@
       "bytes": 6024653,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9568293169977551,
+      "bytes": 2417450,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -310,3 +310,4 @@
 {"step": 268100, "loss": 0.5864941235631704, "lr": 0.0003, "grad_norm": 0.14459926591310038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39241925666, "tokens_per_second": 149693.1115663491, "data_wait_s": 1.1463998816907406e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268200, "loss": 0.540740785188973, "lr": 0.0003, "grad_norm": 0.14683682745459853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39256643236, "tokens_per_second": 149615.4632257869, "data_wait_s": 1.343500116490759e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268300, "loss": 0.6001504268497229, "lr": 0.0003, "grad_norm": 0.16388218285261316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39271157663, "tokens_per_second": 148489.45665295143, "data_wait_s": 1.3158998626749963e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 268400, "loss": 0.6030313111841679, "lr": 0.0003, "grad_norm": 0.15376708163393907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39285573765, "tokens_per_second": 147003.52498788704, "data_wait_s": 1.2368000170681626e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4e4e59a26cb4277edb8ab381dc45cb4ead810ca429bd7973e232a4a4ee18313
-size 665679
+oid sha256:940dffb627a7d131397f266b0fb62284b39e94fa04fe251cca6ef568abe5018f
+size 667797