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": 77258.21737409224,
-      "required_pages_per_second_upper_estimate": 0.005344902741701736
+      "required_token_positions_per_second": 75903.59190667889,
+      "required_pages_per_second_upper_estimate": 0.005251186608701952
     },
     {
       "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": 9657.27717176153,
-      "required_pages_per_second_upper_estimate": 0.004043300284936901
+      "required_token_positions_per_second": 9487.948988334861,
+      "required_pages_per_second_upper_estimate": 0.003972406110510714
     },
     {
       "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": 19314.55434352306,
-      "required_pages_per_second_upper_estimate": 0.004022712310704902
+      "required_token_positions_per_second": 18975.897976669723,
+      "required_pages_per_second_upper_estimate": 0.0039521791204583274
     },
     {
       "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": 28971.83151528459,
-      "required_pages_per_second_upper_estimate": 0.0017368170619160173
+      "required_token_positions_per_second": 28463.84696500458,
+      "required_pages_per_second_upper_estimate": 0.0017063641637742276
     },
     {
       "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": 28971.83151528459,
-      "required_pages_per_second_upper_estimate": 0.002252432466694141
+      "required_token_positions_per_second": 28463.84696500458,
+      "required_pages_per_second_upper_estimate": 0.0022129389022977697
     },
     {
       "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": 28971.83151528459,
-      "required_pages_per_second_upper_estimate": 0.37161954714901796
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28463.84696500458,
+      "required_pages_per_second_upper_estimate": 0.36510366676934086
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.11458330740617839,
-    "producer_p95_seconds": 0.3135101330044563,
-    "gpu_update_seconds": 0.9543062538318925,
-    "producer_batches_per_second": 8.727274702022429,
-    "gpu_batches_per_second": 1.04788163756093,
-    "producer_headroom": 8.328492827048871,
+    "producer_batch_seconds": 0.5730413844145801,
+    "producer_p95_seconds": 0.3138444790020003,
+    "gpu_update_seconds": 0.971337431443907,
+    "producer_batches_per_second": 1.7450746616173307,
+    "gpu_batches_per_second": 1.0295083537689735,
+    "producer_headroom": 1.6950563394932232,
     "sustainable": true,
-    "buffer_seconds": 60.121293991409225,
+    "buffer_seconds": 60.22292074952224,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323457.71941785753,
-    "upload_bytes_per_second": 4986252.333942324,
+    "generation_bytes_per_second": 283836.7356223343,
+    "upload_bytes_per_second": 4929189.991331254,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.40649687673373,
+    "estimated_full_shards_per_hour": 60.90475608351251,
     "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": 1.4667487440019613,
-      "bytes": 7634603,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.2097738270022091,
       "bytes": 9815120,
@@ -562,6 +556,12 @@
       "bytes": 7801020,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.0194173400013824,
+      "bytes": 489708,
+      "pages": 17,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -418,3 +418,4 @@
 {"step": 926800, "loss": 0.5488107800483704, "lr": 3e-05, "grad_norm": 0.35696342018321464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135657183960, "tokens_per_second": 152605.09608179412, "data_wait_s": 1.4655000995844603e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926900, "loss": 0.5539249181747437, "lr": 3e-05, "grad_norm": 0.3245176664790011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135671791690, "tokens_per_second": 153431.22632042432, "data_wait_s": 1.327700010733679e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927000, "loss": 0.5282608270645142, "lr": 3e-05, "grad_norm": 0.3187059187752664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135686554750, "tokens_per_second": 146545.1891244505, "data_wait_s": 1.5465004253201187e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 927100, "loss": 0.5215455889701843, "lr": 3e-05, "grad_norm": 0.33811687711354393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135701278418, "tokens_per_second": 153998.26457812803, "data_wait_s": 1.2737000361084938e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91c303d3b383ca6732c3e62911c5a488a9935708e774973ee815f1486f071f83
-size 895989
+oid sha256:5637d9bc44f8997c883bc00b42792918225e07932cb0dac57a1cdabc75bdbfa5
+size 898107