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": 77066.9791592362,
-      "required_pages_per_second_upper_estimate": 0.005331672438264202
+      "required_token_positions_per_second": 76067.72994170256,
+      "required_pages_per_second_upper_estimate": 0.005262542058817602
     },
     {
       "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": 9633.372394904525,
-      "required_pages_per_second_upper_estimate": 0.004033291854055378
+      "required_token_positions_per_second": 9508.46624271282,
+      "required_pages_per_second_upper_estimate": 0.003980996256469773
     },
     {
       "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": 19266.74478980905,
-      "required_pages_per_second_upper_estimate": 0.004012754841488992
+      "required_token_positions_per_second": 19016.93248542564,
+      "required_pages_per_second_upper_estimate": 0.003960725526479417
     },
     {
       "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": 28900.117184713574,
-      "required_pages_per_second_upper_estimate": 0.0017325178972997274
+      "required_token_positions_per_second": 28525.398728138458,
+      "required_pages_per_second_upper_estimate": 0.0017100540979900032
     },
     {
       "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": 28900.117184713574,
-      "required_pages_per_second_upper_estimate": 0.002246856992930248
+      "required_token_positions_per_second": 28525.398728138458,
+      "required_pages_per_second_upper_estimate": 0.002217724281143835
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28900.117184713574,
-      "required_pages_per_second_upper_estimate": 0.3706996727173019
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28525.398728138458,
+      "required_pages_per_second_upper_estimate": 0.36589318669768806
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09173641297659074,
-    "producer_p95_seconds": 0.1515816320002159,
-    "gpu_update_seconds": 0.9566743215361123,
-    "producer_batches_per_second": 10.900796832498559,
-    "gpu_batches_per_second": 1.04528780326655,
-    "producer_headroom": 10.42851241393356,
+    "producer_batch_seconds": 0.3812643923827892,
+    "producer_p95_seconds": 0.15603487699991092,
+    "gpu_update_seconds": 0.9692414911882384,
+    "producer_batches_per_second": 2.622851805672953,
+    "gpu_batches_per_second": 1.0317346183499154,
+    "producer_headroom": 2.542176795296217,
     "sustainable": true,
-    "buffer_seconds": 60.270482256775075,
+    "buffer_seconds": 60.09297245367078,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352983.54611674964,
-    "upload_bytes_per_second": 5676419.486963453,
+    "generation_bytes_per_second": 324531.89493862295,
+    "upload_bytes_per_second": 5832761.838987507,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.74205196024768,
+    "estimated_full_shards_per_hour": 69.63698993796365,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 6102397,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0165042769999673,
+      "bytes": 8195972,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 602600, "loss": 0.5469270944595337, "lr": 0.0003, "grad_norm": 0.15803339422789744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88202377633, "tokens_per_second": 151909.8445450612, "data_wait_s": 1.2728999990940792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 602700, "loss": 0.6067249774932861, "lr": 0.0003, "grad_norm": 0.14774953975413252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88217111566, "tokens_per_second": 153377.78396283227, "data_wait_s": 1.471799987484701e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 602800, "loss": 0.6329813003540039, "lr": 0.0003, "grad_norm": 0.15991983956735892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88231758455, "tokens_per_second": 151709.75645617614, "data_wait_s": 1.4028999885340454e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 602900, "loss": 0.5468085408210754, "lr": 0.0003, "grad_norm": 0.148908295804209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88246500487, "tokens_per_second": 153463.02942106282, "data_wait_s": 1.2954999874636997e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4caf417e63d1f2b1ec4a08acd1b99be0e0450c3df81eb666321d38493501b52
-size 59469
+oid sha256:bf0e00206533a77a8d0ed07836dc14280ea35c94aa86cab53483463b0089165f
+size 61587