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": 77152.88204867867,
-      "required_pages_per_second_upper_estimate": 0.005337615399477997
+      "required_token_positions_per_second": 76915.46962991485,
+      "required_pages_per_second_upper_estimate": 0.005321190657475209
     },
     {
       "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": 9644.110256084834,
-      "required_pages_per_second_upper_estimate": 0.004037787572299534
+      "required_token_positions_per_second": 9614.433703739356,
+      "required_pages_per_second_upper_estimate": 0.004025362619549366
     },
     {
       "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": 19288.220512169668,
-      "required_pages_per_second_upper_estimate": 0.004017227668103825
+      "required_token_positions_per_second": 19228.867407478712,
+      "required_pages_per_second_upper_estimate": 0.004004865981643327
     },
     {
       "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": 28932.3307682545,
-      "required_pages_per_second_upper_estimate": 0.0017344490524457117
+      "required_token_positions_per_second": 28843.301111218065,
+      "required_pages_per_second_upper_estimate": 0.0017291118604467926
     },
     {
       "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": 28932.3307682545,
-      "required_pages_per_second_upper_estimate": 0.0022493614573579758
+      "required_token_positions_per_second": 28843.301111218065,
+      "required_pages_per_second_upper_estimate": 0.002242439793123469
     },
     {
       "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": 28932.3307682545,
-      "required_pages_per_second_upper_estimate": 0.3711128739787137
+      "required_token_positions_per_second": 28843.301111218065,
+      "required_pages_per_second_upper_estimate": 0.3699708971308483
     }
   ],
   "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": 0.0804450490546742,
-    "producer_p95_seconds": 0.2308787679994566,
-    "gpu_update_seconds": 0.9556091495516943,
-    "producer_batches_per_second": 12.430845797860766,
-    "gpu_batches_per_second": 1.0464529357730938,
-    "producer_headroom": 11.87902998110198,
+    "producer_batch_seconds": 0.11052786946089554,
+    "producer_p95_seconds": 0.21978281100018648,
+    "gpu_update_seconds": 0.958558796491114,
+    "producer_batches_per_second": 9.047491866780236,
+    "gpu_batches_per_second": 1.0432328237564406,
+    "producer_headroom": 8.672552915084005,
     "sustainable": true,
-    "buffer_seconds": 60.203376421756744,
+    "buffer_seconds": 60.38920417894018,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465055.1865316767,
-    "upload_bytes_per_second": 2186944.7682679635,
+    "generation_bytes_per_second": 489061.3728072803,
+    "upload_bytes_per_second": 2164942.1149037983,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.79001709902501,
+    "estimated_full_shards_per_hour": 104.94118583835417,
     "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": 1.4639023600002474,
-      "bytes": 7690441,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 0.9298561730001893,
       "bytes": 6431657,
@@ -562,6 +556,12 @@
       "bytes": 6318440,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1342329689996404,
+      "bytes": 2581964,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -156,3 +156,4 @@
 {"step": 660600, "loss": 0.522675633430481, "lr": 0.0001, "grad_norm": 0.19461194593324185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96694521168, "tokens_per_second": 152697.3829239485, "data_wait_s": 1.2554000932141207e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 660700, "loss": 0.5982367396354675, "lr": 0.0001, "grad_norm": 0.20127042855920593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96709222604, "tokens_per_second": 153484.15191274587, "data_wait_s": 8.015000275918283e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 660800, "loss": 0.5168361067771912, "lr": 0.0001, "grad_norm": 0.19015320190819313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96723759709, "tokens_per_second": 151539.75673439808, "data_wait_s": 1.0210000255028717e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 660900, "loss": 0.5548530220985413, "lr": 0.0001, "grad_norm": 0.18112856739384112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96738195214, "tokens_per_second": 151036.16568345609, "data_wait_s": 8.747998435865156e-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:ebcc164c79c745607084a3a8e1476adb5fe976b6ebcabea6990864a24b659fd6
-size 337158
+oid sha256:dd54a8cfca4eb6519a9332472b85f632365ee0ffeeff0618f143130acc163282
+size 339276