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": 76879.19937554622,
-      "required_pages_per_second_upper_estimate": 0.005318681397119401
+      "required_token_positions_per_second": 76437.5875524402,
+      "required_pages_per_second_upper_estimate": 0.005288129666516325
     },
     {
       "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": 9609.899921943277,
-      "required_pages_per_second_upper_estimate": 0.004023464419787477
+      "required_token_positions_per_second": 9554.698444055026,
+      "required_pages_per_second_upper_estimate": 0.004000352713733607
     },
     {
       "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": 19219.799843886554,
-      "required_pages_per_second_upper_estimate": 0.004002977447274813
+      "required_token_positions_per_second": 19109.39688811005,
+      "required_pages_per_second_upper_estimate": 0.003979983423108303
     },
     {
       "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": 28829.69976582983,
-      "required_pages_per_second_upper_estimate": 0.00172829648055882
+      "required_token_positions_per_second": 28664.095332165078,
+      "required_pages_per_second_upper_estimate": 0.001718368747623947
     },
     {
       "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": 28829.69976582983,
-      "required_pages_per_second_upper_estimate": 0.0022413823483455296
+      "required_token_positions_per_second": 28664.095332165078,
+      "required_pages_per_second_upper_estimate": 0.002228507332045017
     },
     {
       "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": 28829.69976582983,
-      "required_pages_per_second_upper_estimate": 0.3697964336761949
+      "required_token_positions_per_second": 28664.095332165078,
+      "required_pages_per_second_upper_estimate": 0.36767223781333075
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11003331969527608,
-    "producer_p95_seconds": 0.3217869329964742,
-    "gpu_update_seconds": 0.9590110276753409,
-    "producer_batches_per_second": 9.088156230943305,
-    "gpu_batches_per_second": 1.0427408769469702,
-    "producer_headroom": 8.715642046710991,
+    "producer_batch_seconds": 0.12034655742962741,
+    "producer_p95_seconds": 0.32297176199790556,
+    "gpu_update_seconds": 0.9645516343568367,
+    "producer_batches_per_second": 8.309336148520488,
+    "gpu_batches_per_second": 1.0367511332524986,
+    "producer_headroom": 8.01478376247578,
     "sustainable": true,
-    "buffer_seconds": 60.41769474354648,
+    "buffer_seconds": 60.76675296448071,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 238817.53026113464,
+    "generation_bytes_per_second": 264467.8819513984,
     "upload_bytes_per_second": 4724037.1197781535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.24468260646371,
+    "estimated_full_shards_per_hour": 56.74865096956695,
     "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": [

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -405,3 +405,4 @@
 {"step": 760500, "loss": 0.5541071891784668, "lr": 5e-05, "grad_norm": 0.2431569137163541, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111314904726, "tokens_per_second": 151159.36517041238, "data_wait_s": 1.4037999790161848e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760600, "loss": 0.5564252734184265, "lr": 5e-05, "grad_norm": 0.24219025165694613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111329597564, "tokens_per_second": 151954.04795665306, "data_wait_s": 1.232800423167646e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760700, "loss": 0.45216894149780273, "lr": 5e-05, "grad_norm": 0.24528988115250652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111344364855, "tokens_per_second": 152424.21800224096, "data_wait_s": 1.9842998881358653e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 760800, "loss": 0.535881519317627, "lr": 5e-05, "grad_norm": 0.24302539573863297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111358936214, "tokens_per_second": 150784.03338170476, "data_wait_s": 1.6279002011287957e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33ef65f9b2c3597ed402568650b7641a0be68bc448ec7e0546ade4e4f7f007c7
-size 868455
+oid sha256:f63bfa83b9f0a11efeaa9cd3798c366ffc09cc15e6071c61b6d97a6d9d5aa2a3
+size 870573