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": 76937.50859922016,
-      "required_pages_per_second_upper_estimate": 0.0053227153645092045
+      "required_token_positions_per_second": 76525.07455078278,
+      "required_pages_per_second_upper_estimate": 0.005294182220059468
     },
     {
       "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": 9617.18857490252,
-      "required_pages_per_second_upper_estimate": 0.004026516026577131
+      "required_token_positions_per_second": 9565.634318847848,
+      "required_pages_per_second_upper_estimate": 0.004004931336142327
     },
     {
       "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": 19234.37714980504,
-      "required_pages_per_second_upper_estimate": 0.004006013515668225
+      "required_token_positions_per_second": 19131.268637695695,
+      "required_pages_per_second_upper_estimate": 0.003984538731750168
     },
     {
       "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": 28851.565724707558,
-      "required_pages_per_second_upper_estimate": 0.0017296073114061537
+      "required_token_positions_per_second": 28696.902956543545,
+      "required_pages_per_second_upper_estimate": 0.0017203355146110941
     },
     {
       "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": 28851.565724707558,
-      "required_pages_per_second_upper_estimate": 0.002243082330470084
+      "required_token_positions_per_second": 28696.902956543545,
+      "required_pages_per_second_upper_estimate": 0.002231057980534954
     },
     {
       "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": 28851.565724707558,
-      "required_pages_per_second_upper_estimate": 0.3700769067188409
+      "required_token_positions_per_second": 28696.902956543545,
+      "required_pages_per_second_upper_estimate": 0.36809305879277515
     }
   ],
   "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": 0.13055381181226267,
-    "producer_p95_seconds": 0.30772295399947325,
-    "gpu_update_seconds": 0.9582842145833055,
-    "producer_batches_per_second": 7.659676773268078,
-    "gpu_batches_per_second": 1.0435317464086935,
-    "producer_headroom": 7.340147340633188,
+    "producer_batch_seconds": 0.08949301910951135,
+    "producer_p95_seconds": 0.29939615599869285,
+    "gpu_update_seconds": 0.963448914395678,
+    "producer_batches_per_second": 11.174055920231208,
+    "gpu_batches_per_second": 1.0379377516110946,
+    "producer_headroom": 10.765632045743358,
     "sustainable": true,
-    "buffer_seconds": 60.37190551874824,
+    "buffer_seconds": 60.69728160692772,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400071.0422222372,
-    "upload_bytes_per_second": 4938723.990978582,
+    "generation_bytes_per_second": 399990.19049658516,
+    "upload_bytes_per_second": 4918713.34758047,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.84593248367631,
+    "estimated_full_shards_per_hour": 85.82858358548323,
     "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": 0.836775318999571,
-      "bytes": 1105976,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.2929117179992318,
       "bytes": 11923823,
@@ -562,6 +556,12 @@
       "bytes": 13187614,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.1867789399984758,
+      "bytes": 1047762,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 794100, "loss": 0.5055521726608276, "lr": 3e-05, "grad_norm": 0.258596852103173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116233781402, "tokens_per_second": 153481.87996062965, "data_wait_s": 9.018003765959293e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 794200, "loss": 0.5263431072235107, "lr": 3e-05, "grad_norm": 0.2563706211314849, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116248491627, "tokens_per_second": 152643.05332925252, "data_wait_s": 1.2525997590273619e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 794300, "loss": 0.5353758335113525, "lr": 3e-05, "grad_norm": 0.2582250826480756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116263171118, "tokens_per_second": 152911.1696900838, "data_wait_s": 9.776005754247308e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 794400, "loss": 0.5692605376243591, "lr": 3e-05, "grad_norm": 0.2803299069318174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116277915773, "tokens_per_second": 153034.29447523592, "data_wait_s": 1.064100069925189e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e38c098903b1e3a10d28519dcd1701b656569d2751e2c6cabcff9d2f8affbaeb
-size 731337
+oid sha256:014ad342e9345efc8885c89504eee961f49e19b818d85ad510021ffb6db938a4
+size 733455