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": 75609.30605616777,
-      "required_pages_per_second_upper_estimate": 0.005230827230726343
+      "required_token_positions_per_second": 76146.27594917406,
+      "required_pages_per_second_upper_estimate": 0.005267976053866327
     },
     {
       "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": 9451.163257020971,
-      "required_pages_per_second_upper_estimate": 0.00395700469298301
+      "required_token_positions_per_second": 9518.284493646757,
+      "required_pages_per_second_upper_estimate": 0.003985106953107419
     },
     {
       "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": 18902.326514041943,
-      "required_pages_per_second_upper_estimate": 0.003936856125002904
+      "required_token_positions_per_second": 19036.568987293515,
+      "required_pages_per_second_upper_estimate": 0.003964815291969117
     },
     {
       "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": 28353.489771062912,
-      "required_pages_per_second_upper_estimate": 0.0016997484185031029
+      "required_token_positions_per_second": 28554.85348094027,
+      "required_pages_per_second_upper_estimate": 0.0017118198654456683
     },
     {
       "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": 28353.489771062912,
-      "required_pages_per_second_upper_estimate": 0.0022043591158788013
+      "required_token_positions_per_second": 28554.85348094027,
+      "required_pages_per_second_upper_estimate": 0.0022200142586164106
     },
     {
       "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": 28353.489771062912,
-      "required_pages_per_second_upper_estimate": 0.36368812317778004
+      "required_token_positions_per_second": 28554.85348094027,
+      "required_pages_per_second_upper_estimate": 0.36627100064058016
     }
   ],
   "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": 0.1110065482116056,
-    "producer_p95_seconds": 0.2992891920002876,
-    "gpu_update_seconds": 0.9751180621235935,
-    "producer_batches_per_second": 9.008477572816297,
-    "gpu_batches_per_second": 1.0255168464649491,
-    "producer_headroom": 8.78432919348848,
+    "producer_batch_seconds": 0.13803465728142328,
+    "producer_p95_seconds": 0.2903211709999596,
+    "gpu_update_seconds": 0.9682417042852076,
+    "producer_batches_per_second": 7.244557415470036,
+    "gpu_batches_per_second": 1.0327999667585457,
+    "producer_headroom": 7.014482618746746,
     "sustainable": true,
-    "buffer_seconds": 60.4573198516628,
+    "buffer_seconds": 60.03098566568287,
     "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": 472523.53523948876,
-    "upload_bytes_per_second": 5047737.025873366,
+    "generation_bytes_per_second": 470289.3936351328,
+    "upload_bytes_per_second": 5051073.799909289,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.3925508774614,
+    "estimated_full_shards_per_hour": 100.9131560973214,
     "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": 1.1213059549991158,
-      "bytes": 6142168,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3281040260008012,
       "bytes": 876378,
@@ -562,6 +556,12 @@
       "bytes": 11656046,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.130367224002839,
+      "bytes": 6468621,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -340,3 +340,4 @@
 {"step": 754000, "loss": 0.563613772392273, "lr": 5e-05, "grad_norm": 0.2313926614011883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110364629186, "tokens_per_second": 151548.63714057222, "data_wait_s": 1.344000338576734e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 754100, "loss": 0.5629440546035767, "lr": 5e-05, "grad_norm": 0.2371682280083104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110379439431, "tokens_per_second": 152610.93935221995, "data_wait_s": 1.5237004845403135e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 754200, "loss": 0.5593996644020081, "lr": 5e-05, "grad_norm": 0.23872154556980332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110394023411, "tokens_per_second": 150613.9667814889, "data_wait_s": 1.1531003110576421e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 754300, "loss": 0.5498316884040833, "lr": 5e-05, "grad_norm": 0.2421564901006812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110408529030, "tokens_per_second": 149676.33542256468, "data_wait_s": 1.0902003850787878e-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 sha256:2dc38153b03c36b5b870dc13d875e82a58d0f345bd6df3d8c4b41d3a9ac0fadf
-size 729219
+oid sha256:6d13a22797e31f4cf87745a9c8b3c3665134d091f005fe641956d8a37ccdafde
+size 731337