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": 76515.5646239516,
-      "required_pages_per_second_upper_estimate": 0.005293524301255217
+      "required_token_positions_per_second": 76187.17829520178,
+      "required_pages_per_second_upper_estimate": 0.005270805773071044
     },
     {
       "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": 9564.44557799395,
-      "required_pages_per_second_upper_estimate": 0.0040044336351705324
+      "required_token_positions_per_second": 9523.397286900223,
+      "required_pages_per_second_upper_estimate": 0.00398724757287538
     },
     {
       "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": 19128.8911559879,
-      "required_pages_per_second_upper_estimate": 0.003984043565008844
+      "required_token_positions_per_second": 19046.794573800446,
+      "required_pages_per_second_upper_estimate": 0.003966945011971661
     },
     {
       "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": 28693.33673398185,
-      "required_pages_per_second_upper_estimate": 0.0017201217250138272
+      "required_token_positions_per_second": 28570.191860700666,
+      "required_pages_per_second_upper_estimate": 0.0017127393778919543
     },
     {
       "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": 28693.33673398185,
-      "required_pages_per_second_upper_estimate": 0.0022307807224169374
+      "required_token_positions_per_second": 28570.191860700666,
+      "required_pages_per_second_upper_estimate": 0.002221206750176379
     },
     {
       "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": 28693.33673398185,
-      "required_pages_per_second_upper_estimate": 0.3680473151188652
+      "required_token_positions_per_second": 28570.191860700666,
+      "required_pages_per_second_upper_estimate": 0.36646774490707745
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1079776353594184,
-    "producer_p95_seconds": 0.24411641499864345,
-    "gpu_update_seconds": 0.9635686590349094,
-    "producer_batches_per_second": 9.261177063855516,
-    "gpu_batches_per_second": 1.0378087649733019,
-    "producer_headroom": 8.923779964504119,
+    "producer_batch_seconds": 0.08000739890631792,
+    "producer_p95_seconds": 0.23222541799987084,
+    "gpu_update_seconds": 0.9677218877214061,
+    "producer_batches_per_second": 12.49884402779945,
+    "gpu_batches_per_second": 1.0333547403320553,
+    "producer_headroom": 12.095404936917507,
     "sustainable": true,
-    "buffer_seconds": 60.70482551919929,
+    "buffer_seconds": 60.96647892644859,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443389.0730257147,
-    "upload_bytes_per_second": 5640585.802404035,
+    "generation_bytes_per_second": 461033.74658048444,
+    "upload_bytes_per_second": 5587490.42859073,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.14097350195486,
+    "estimated_full_shards_per_hour": 98.92710969983005,
     "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": 1.3184527379999054,
-      "bytes": 7724848,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9524561609996454,
       "bytes": 5747185,
@@ -562,6 +556,12 @@
       "bytes": 10175048,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0568985930003691,
+      "bytes": 2271584,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -146,3 +146,4 @@
 {"step": 411700, "loss": 0.5794970393180847, "lr": 0.0003, "grad_norm": 0.1733984960715895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60262659478, "tokens_per_second": 151426.0750094452, "data_wait_s": 1.756699930410832e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 411800, "loss": 0.6317907571792603, "lr": 0.0003, "grad_norm": 0.16715218412259703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60277173159, "tokens_per_second": 150989.12647197262, "data_wait_s": 1.9148999854223803e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 411900, "loss": 0.5704567432403564, "lr": 0.0003, "grad_norm": 0.16515601156868315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60291764012, "tokens_per_second": 151566.4214560717, "data_wait_s": 8.141001671901904e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 412000, "loss": 0.5596719980239868, "lr": 0.0003, "grad_norm": 0.1739813141639292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60306317165, "tokens_per_second": 150413.91060559504, "data_wait_s": 1.1639000149443746e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0392be92e45c0edd9c5834d94a9bc8ba73e80f3b58fc249c7b363ba34e9ae1a
-size 315978
+oid sha256:52dc805452a72aa3258954eb5013aa64974ce019467969b25c97d1beb95f1a07
+size 318096