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": 77242.74051920265,
-      "required_pages_per_second_upper_estimate": 0.005343832016969223
+      "required_token_positions_per_second": 76405.51506043682,
+      "required_pages_per_second_upper_estimate": 0.005285910817101096
     },
     {
       "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": 9655.342564900331,
-      "required_pages_per_second_upper_estimate": 0.004042490305443302
+      "required_token_positions_per_second": 9550.689382554603,
+      "required_pages_per_second_upper_estimate": 0.003998674203402104
     },
     {
       "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": 19310.685129800662,
-      "required_pages_per_second_upper_estimate": 0.004021906455524554
+      "required_token_positions_per_second": 19101.378765109206,
+      "required_pages_per_second_upper_estimate": 0.00397831345953935
     },
     {
       "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": 28966.02769470099,
-      "required_pages_per_second_upper_estimate": 0.0017364691317339515
+      "required_token_positions_per_second": 28652.06814766381,
+      "required_pages_per_second_upper_estimate": 0.0017176477362775434
     },
     {
       "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": 28966.02769470099,
-      "required_pages_per_second_upper_estimate": 0.002251981245172075
+      "required_token_positions_per_second": 28652.06814766381,
+      "required_pages_per_second_upper_estimate": 0.0022275722713520373
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28966.02769470099,
-      "required_pages_per_second_upper_estimate": 0.3715451019702286
+      "required_token_positions_per_second": 28652.06814766381,
+      "required_pages_per_second_upper_estimate": 0.36751796600433306
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11002035905474372,
-    "producer_p95_seconds": 0.27520782799911103,
-    "gpu_update_seconds": 0.9544974648028073,
-    "producer_batches_per_second": 9.089226835757025,
-    "gpu_batches_per_second": 1.0476717192817198,
-    "producer_headroom": 8.675643971747721,
+    "producer_batch_seconds": 0.12263013957002045,
+    "producer_p95_seconds": 0.26608256599865854,
+    "gpu_update_seconds": 0.9649565210270632,
+    "producer_batches_per_second": 8.15460215169217,
+    "gpu_batches_per_second": 1.0363161222389978,
+    "producer_headroom": 7.868836522656681,
     "sustainable": true,
-    "buffer_seconds": 60.13334028257686,
+    "buffer_seconds": 60.79226082470498,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 603767.2841028727,
-    "upload_bytes_per_second": 5368129.427869143,
+    "generation_bytes_per_second": 616404.8608709782,
+    "upload_bytes_per_second": 5317205.715010796,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.77720209271735,
+    "estimated_full_shards_per_hour": 66.13306698606972,
     "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": 1.1316358820004098,
-      "bytes": 5807643,
-      "pages": 41,
-      "shards": 2
-    },
     {
       "seconds": 1.0283135649988253,
       "bytes": 5734251,
@@ -562,6 +556,12 @@
       "bytes": 5824332,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3192006099998252,
+      "bytes": 2683646,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -254,3 +254,4 @@
 {"step": 670400, "loss": 0.4927593171596527, "lr": 0.0001, "grad_norm": 0.1783781125813963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98129679618, "tokens_per_second": 151830.30857302772, "data_wait_s": 1.2854001397499815e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 670500, "loss": 0.5398595333099365, "lr": 0.0001, "grad_norm": 0.18922684276234444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98144384948, "tokens_per_second": 153014.43667492302, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 670600, "loss": 0.535764753818512, "lr": 0.0001, "grad_norm": 0.18723919860453841, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98158886266, "tokens_per_second": 150746.4327942166, "data_wait_s": 1.76030007423833e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 670700, "loss": 0.5527182817459106, "lr": 0.0001, "grad_norm": 0.19289207468098582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98173372423, "tokens_per_second": 151383.34119689272, "data_wait_s": 9.703002433525398e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f08d07f13622c675c1ea35b728f97cdfaafd280b4b48f3f341135260311982a
-size 546288
+oid sha256:421bffb4d9c19aed9e7f2891e137a2583713dbb34a483ca68a1aceb2e44fbad1
+size 548406