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": 75367.00913212574,
-      "required_pages_per_second_upper_estimate": 0.0065326213356343655
+      "required_token_positions_per_second": 75568.0566287301,
+      "required_pages_per_second_upper_estimate": 0.006550047623089811
     },
     {
       "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": 9420.876141515717,
-      "required_pages_per_second_upper_estimate": 0.003758128347501084
+      "required_token_positions_per_second": 9446.007078591263,
+      "required_pages_per_second_upper_estimate": 0.0037681534540415125
     },
     {
       "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": 18841.752283031434,
-      "required_pages_per_second_upper_estimate": 0.004358707787164807
+      "required_token_positions_per_second": 18892.014157182526,
+      "required_pages_per_second_upper_estimate": 0.004370334987170889
     },
     {
       "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": 28262.62842454715,
-      "required_pages_per_second_upper_estimate": 0.0017210457885034572
+      "required_token_positions_per_second": 28338.021235773787,
+      "required_pages_per_second_upper_estimate": 0.001725636815151649
     },
     {
       "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": 28262.62842454715,
-      "required_pages_per_second_upper_estimate": 0.0021757353720844434
+      "required_token_positions_per_second": 28338.021235773787,
+      "required_pages_per_second_upper_estimate": 0.0021815393193933288
     },
     {
       "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": 28262.62842454715,
-      "required_pages_per_second_upper_estimate": 0.35092725610026637
+      "required_token_positions_per_second": 28338.021235773787,
+      "required_pages_per_second_upper_estimate": 0.3518633824921935
     }
   ],
   "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": 0.0604805084686717,
-    "producer_p95_seconds": 0.20177922799848602,
-    "gpu_update_seconds": 0.9782529630537362,
-    "producer_batches_per_second": 16.534252527291333,
-    "gpu_batches_per_second": 1.0222304841054382,
-    "producer_headroom": 16.174681526701473,
+    "producer_batch_seconds": 0.0820072297733958,
+    "producer_p95_seconds": 0.222282536999046,
+    "gpu_update_seconds": 0.9756503381081983,
+    "producer_batches_per_second": 12.194046826886144,
+    "gpu_batches_per_second": 1.024957365298531,
+    "producer_headroom": 11.89712590955867,
     "sustainable": true,
-    "buffer_seconds": 60.65168370933164,
+    "buffer_seconds": 60.490320962708296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263954.33021359734,
-    "upload_bytes_per_second": 4042786.830392722,
+    "generation_bytes_per_second": 357826.5870109013,
+    "upload_bytes_per_second": 4031612.6683875076,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.63845472150746,
+    "estimated_full_shards_per_hour": 76.78125579591064,
     "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": 0.7353289520001454,
-      "bytes": 1217005,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.1706305770003382,
       "bytes": 7901550,
@@ -562,6 +556,12 @@
       "bytes": 12190710,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.9488432640009705,
+      "bytes": 921326,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -105,3 +105,4 @@
 {"step": 325500, "loss": 0.6061256788671017, "lr": 0.0003, "grad_norm": 0.1477813010888525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47644090091, "tokens_per_second": 145756.85641014134, "data_wait_s": 1.6502001017215662e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325600, "loss": 0.6151839569211006, "lr": 0.0003, "grad_norm": 0.15870963723315856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47658801353, "tokens_per_second": 149701.9202591019, "data_wait_s": 1.384799907100387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325700, "loss": 0.6090668737888336, "lr": 0.0003, "grad_norm": 0.1644471172121183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47673575407, "tokens_per_second": 150253.5453183789, "data_wait_s": 1.8262000594404526e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 325800, "loss": 0.584720553830266, "lr": 0.0003, "grad_norm": 0.14197307855976757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47688314798, "tokens_per_second": 149828.13113493362, "data_wait_s": 1.1918999007320963e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:184eb50f7359c1a8d354cd129c6bae8b29a37f975591f9fb257db33837d09496
-size 228357
+oid sha256:05389ec014fea2e651f707d732150f2321bdddbf2ef9509068f3741b60e5b1aa
+size 230475