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": 77004.71825303738,
-      "required_pages_per_second_upper_estimate": 0.005327365084308162
+      "required_token_positions_per_second": 76373.26390214733,
+      "required_pages_per_second_upper_estimate": 0.005283679607137632
     },
     {
       "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": 9625.589781629673,
-      "required_pages_per_second_upper_estimate": 0.004030033436396643
+      "required_token_positions_per_second": 9546.657987768416,
+      "required_pages_per_second_upper_estimate": 0.003996986342590224
     },
     {
       "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": 19251.179563259346,
-      "required_pages_per_second_upper_estimate": 0.004009513015281316
+      "required_token_positions_per_second": 19093.315975536832,
+      "required_pages_per_second_upper_estimate": 0.003976634193101484
     },
     {
       "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": 28876.76934488902,
-      "required_pages_per_second_upper_estimate": 0.0017311182299523326
+      "required_token_positions_per_second": 28639.97396330525,
+      "required_pages_per_second_upper_estimate": 0.001716922708391998
     },
     {
       "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": 28876.76934488902,
-      "required_pages_per_second_upper_estimate": 0.0022450418010801832
+      "required_token_positions_per_second": 28639.97396330525,
+      "required_pages_per_second_upper_estimate": 0.002226632001714854
     },
     {
       "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": 28876.76934488902,
-      "required_pages_per_second_upper_estimate": 0.3704001916969898
+      "required_token_positions_per_second": 28639.97396330525,
+      "required_pages_per_second_upper_estimate": 0.3673628347931049
     }
   ],
   "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": 0.09654522809415766,
-    "producer_p95_seconds": 0.35309308300202247,
-    "gpu_update_seconds": 0.9574478249206744,
-    "producer_batches_per_second": 10.357839737296286,
-    "gpu_batches_per_second": 1.0444433356803031,
-    "producer_headroom": 9.917091127351258,
+    "producer_batch_seconds": 0.4291288818903354,
+    "producer_p95_seconds": 0.35406761200283654,
+    "gpu_update_seconds": 0.9653640061064229,
+    "producer_batches_per_second": 2.330302252309253,
+    "gpu_batches_per_second": 1.0358786879088993,
+    "producer_headroom": 2.249589917728081,
     "sustainable": true,
-    "buffer_seconds": 60.31921297000248,
+    "buffer_seconds": 60.81793238470465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301164.44499186124,
-    "upload_bytes_per_second": 4261418.34341481,
+    "generation_bytes_per_second": 283354.6031974931,
+    "upload_bytes_per_second": 4231794.333385845,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.62287914578322,
+    "estimated_full_shards_per_hour": 60.80130168861003,
     "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": 1.7729905640007928,
-      "bytes": 7619842,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.2987675089971162,
       "bytes": 1223490,
@@ -562,6 +556,12 @@
       "bytes": 11724055,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.9323788800029433,
+      "bytes": 1111114,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -405,3 +405,4 @@
 {"step": 840500, "loss": 0.5224630236625671, "lr": 3e-05, "grad_norm": 0.29421269579808434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123025473590, "tokens_per_second": 152329.25091578267, "data_wait_s": 1.0652998753357679e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 840600, "loss": 0.5223937034606934, "lr": 3e-05, "grad_norm": 0.283010003341772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123040191935, "tokens_per_second": 152369.94663505742, "data_wait_s": 1.1866002751048654e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 840700, "loss": 0.5038002729415894, "lr": 3e-05, "grad_norm": 0.2786213606761156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123054808705, "tokens_per_second": 151234.74413240745, "data_wait_s": 1.2336000509094447e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 840800, "loss": 0.5980193018913269, "lr": 3e-05, "grad_norm": 0.32692752093060673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123069241731, "tokens_per_second": 150399.62103053424, "data_wait_s": 1.2342003174126148e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3c2af869882c416d37c6d16f8ddb870765c4dc28c7a49081ddb311862246417
-size 868455
+oid sha256:005e35d2da8baad36aef9e96e58a90903007d3b1e5b5a4ebc4e94f166200a581
+size 870573